FINAL REPORT

• Document Type: CREST
• Collection: NGA Records (Formerly NIMA)
• Document Number (FOIA) /ESDN (CREST): CIA-RDP78B04770A002300020010-5
• Release Decision: RIPPUB
• Original Classification: C
• Document Page Count: 280
• Document Creation Date: December 28, 2016
• Document Release Date: March 30, 2005
• Sequence Number: 10
• Publication Date: September 30, 1965
• Content Type: LETTER

z
25X1
glop' \
25X1
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Subject: FINAL REPORT
Reference:
DICK S.
Dear Dick:
164-7-26
Copy/of 3
Enclosed are four copies of a report entitled "Automatic
Image Recognition By Coherent Optical Techniques." This
report is the final report on the referenced task. It is our
understanding that you granted permission to release this
report within the Company.
One copy of the report will be furnished to the contracting
officer for his files.
Enclosures
(Four copies of Report No. 65-C-033,
3 September 1965)
Declass Review by NGA.
GROUPS r
l
MIMED
Iii t: iirlic
Approved For Release 201A0W 111WROP781304770A002300020010-514A5'."
DECLASSIFICATION
25X1
25X1
? RECEIPT. FOR CLASSIFIED MATERIAL
OfF CLASSIFIED INFORMATION ON THIS FORM) t:e--"" 619 7 3
Approve(cira katialn005/05/02 : CIA-RDP78B04770A00230002004 0-5
wit.
30 ;3c .tee
TO
Wei, 3,
Washington, D.C.
FROM
INSTRUCTIONS
1. ORIGINAL Of THIS RECEIPT TO 111 SIGNED? BY RECIPIENT AND RETURNED TO
2. DUPLICATE TO BE RETAINED SY RECIPIENT ?
3. TRIPLICATE TO BE RETAINED BY SENDER It4 SUSPENSE FILE
25X1
INOMIxammammm.
? ACCOUNTA.
MLITT
NUMING
IDENTIFICATION
NUMBER
COPY NO.
AND ?
SERIES
CLASS.
?
DESCRIPTION
Per decor/tents, give Pte., pages, dem, *ether. etc.)
164726?
1&2
C
Letter, T. 0. 99, 30 Sept 1965 and attachments.
?
?
?
,?.......------....?,-,...i,
..
CERTIFICATE
I hove personally received the inoterlal'es identified above. I assume full responsibility for the of. hili?fr
Ain., stereo* and tronsmittel el
--'--- -1 "" .....h.11 I.. ??????donei
with exiitind mew* fillillatio"-
SIGNATURE
25
.
?
NAME
Approved For Release 005/05/02 : CIA-RDP78B0477,9A00 Geoeltos
Dot c'e
X1-
///
Approved For Release 2005/05/02 : CIA-RDP78604770A0023 G)0410-5 , 9 9 7 3
14 May 1965
The undersigned acknowledges receipt of the following material:
1 Roll of 9" film (Unclassified)
32 Stripa of 5" film (Secret)
(RADC Control No. 65-S-13611)
25X1
Nam(
Company
Date
itp /5 41---
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approvcd For Rcicasc 2005/05/02 CIA RDP78604770A002300020010-5
CLASS 2
copy
114),11.. ?e.eD r
FIL,6 qg
itieg-.
AUTOMATIC IMAGE RECOGNITION
BY COHERENT OPTICAL TECHNIQUES
by
Report No. 65-C-033
September 3, 1965
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
25X1 Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770 010-5
0
SUBJECT
Optical Image Recognition
_
NO.
65-C-033
DATE
9/3/65
TITLE Automatic Image Recognition by Coherent'
Optical Techniques
II
GOVT. CLASS None
O. PAGES
270
the
JUMMAKT
This report covers a series of studies aimed at
determining the technical feasibility of an automatic
image recognition system employing coherent optical
techniques. The overall study considered the theoretical
limitations, the photographic or image problems, the
equipment design problems and the problems of automating
the recognition system. Twenty-three different image
and system or equipment variables were studied. The
results of this study indicate that image recognition
by spatial filtering is technically feasible under
various conditions considered.
KEY WORDS
Image Recognition, Spatial Filtering, Optical Information
Processing, Optics, Hologram
25X1
25X1
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
TABLE OF CONTENTS
Page
I.
Automatic Image Recognition by Coherent Optical Techniques
A-1-1
2.
Optimum Modulation Transfer Function for a Recognition System
A-2-1
3.
The Use of Phase Modulating Media for Recording Data and for
A-3-1
Recording Spatial Filters
4.
Obtainable Signal-to-Noise Ratios
A-4-1
5.
Information Capacity of a Recognition System
A-5-1
6.
Effect of Image Size
A-6-1
7.
Equipment
B-1-1
8.
The Effect of Aberrations Due to Glass Plates in a Recognition System
B-2-1
9.
Vibration Analysis of 20-Foot Optical Bench
B-3-1
10.
Beam Attenuation Methods
B-4-1
11.
A New Means for Attenuating or Intensifying the Reference Beam
B-5-1
12.
Use of Glass or Film to Make Spatial Filters
B-6-1
13.
Liquid Gate
B-7-1
14.
Test of S.O. 243 Film for Spatial Filters
B-8-1
15.
Automation
C-1-1
16.
Alignment Tolerance of Spatial Filter
C-2-1
17.
Rotation Tolerance of Spatial Filter
C-3-1
18.
Magnification Tolerance of a Spatial Filter
C-4-1
19.
Time Constants of a Scanning System
C-5-1
20.
Recording Materials Study
C-6-1
21.
Feasibility of Real-Time Filter Generation
C-7-1
22.
Television Readout of Recognition Signals
C-8-1
23.
Problems of the Photograph
D-1-1
24.
Packing Density or Effect of Object Size
D-2-1
25.
Effect of Multiple Identical Targets
D-3-1
26.
Multiple Object Filter
D-4-1
27.
Effects of Obscuration-1 Contrast
D-5-1
28.
Effect of Target Aspect
D-6-1
29.
Effect of Shadows
D-7-1
30.
Test of Recognition on Aerial Film-1
E-1-1
31.
Second Roll of Aerial Film
E-2-1
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
25X1
25X1
27X1
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
Automatic Image Recognition by Coherent Optical Techniques
Introduction
It is assumed that the reader is familiar with the principles of optical
processing of information by coherent light. This was described in an earlier
report "The 3-D Hologram Process and Image Recognition by Spatial Filtering" by
report no. 64GL159.*
The theory and mathematics of the subject were described in "Signal Detection by Complex
A-1-1
Spatial Filtering" by
1964, page 139.
IEEE Transactions on Information Theory, April,
The vocabulary of this subject has been drawn from mathematics, information
theory, radar, interferometery and optics, and is confusing to the beginner. All of
the special terms used in this report are defined in the two references given above.
The purpose of the work covered in this report was to determine the feasibility
of designing a completely automatic recognition device to search aerial camera film
and detect the presence of specified objects or "targets" as they are called in this
report. As conceived, the device would be loaded with a roll of film containing
many frames and also one or several spatial filters for the targets of interest. The
equipment would be turned on and would operate automatically until all the film was
processed. The output would consist of a printed record or set of punched cards which
would give the frame numbers on which the targets occurred, which targets were in what
frames, the total number of targets found on the entire film and possibly other
information. Not specifically covered in this study, but considered were such
additional information as the x and y location of the detected targets in the proper
frames. This could be recorded either as a pair of numbers or a photograph which
would indicate the location of the targets in the frame.
* Internal
leport
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Before such a piece of automatic equipment can be designed it is necessary
to know a number of the engineering features of the system and this study was
planned to cover as many of these as could be foreseen in 1964. The study was
divided into four major portions:
A. Theory
B. Equipment
C. Automation
D. Picture Variables
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5 A- l - 3
Section A - Theory
The main problem considered here was to find if there were any theoretical
limitations to the recognition process which would make an automatic device
impractical. Specifically, one point was to find if the large amount of infor-
mation produced by a highly detailed photograph would be too great for the optical
system to handle. Also, it was desired to know if there was a limitation to the
amount of information that could be stored in the spatial filter and if this would
present a real restriction on the process. It was also desired to know whether
theory indicated any difference in performance or limitation to spatial filters
25X1 made by the
recording.
process of instant photography known as Photoplastic
It was also desirable to calculate the expected signal-to-noise ratio which
the process should produce and compare this to actual values.
These problems were separated into five subjects:
A-1 Maximum Information Content of a Spatial Filter.
A-2 Optimum Modulation Transfer Function for a Recognition System.
A-3 Use of a Phase Modulation Media for Recording Data and for
Recording Spatial Filters.
A-4 Obtainable Signal-to-Noise Ratios.
A-5 Information Capacity of a Recognition System.
A-6 Effect of Image Size.
All of the theoretical work was done
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
,ection A-1
INFORMATION CONTENT OF A TWO-BEAM SPATIAL FILTER
A-1-4
Introduction
When spatial filters are used for signal detection in photo-records it is
necessary to determine what information content the filters must have to adequately
carry out the detection process. Generally, the information content of a filter is
closely related to the space-bandwidth product of the filter. This close relation-
ship is valid only, however, if the filter is a low-pass function. If the filter is a
hand-pass function, the information content and the space-bandwidth product may differ
considerably. The two-beam process used to construct complex-valued filters as real
valued functions on a carrier frequency is a band-pass process so that we can expect
this difference to exhibit itself.
In the next paragraph we will define what we mean by information content and
space-bandwidth product. In the following section we will carry out the analysis
necessary to determine these quantities for spatial filters. The final section will
summarize the results and state a few conclusions.
2. Definition of Space-Bandwidth Product and Information Content
The term "space-bandwidth product" (SBP) is the two-dimensional analogy to the
term "time-bandwidth product" in an electronic system. It is the product of the area
oi a given signal and the number of resolution elements that it contains. If the
signal has area A, and is limited in frequency to po, then SBP = A pc)2 . In any
optical system having bandwidth at least as great as pc), the SBP of the signal is
the same in any plane normal to the optical axis. Thus, the SBP of a signal is
dependent only on the area of the signal and the highest frequency that it contains.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-1-5
The term "information content" implies that we know the area of the signal,
its bandwidth, and the number of states that a resolution element can occupy. Note
that in the definition of information content involves the use of the bandwidth of
the signal which is the difference between the highest and lowest frequencies in
the signal. If the signal is a low-pass function, the lowest frequency is zero and
the bandwidth B is simply pc). If the signal is a band-pass function with lowest
frequency pl, then the bandwidth B = p0-p1.
In the noise-free case, a resolution element might occupy any one of N states.
If the probabilities associated with these states are denoted by pi, then we can say
that each element can contain:
II =
pi log2 Pi
bits of information. If all the pi are equal, H = log2 N. The information content
of the filter is then said to be:
IC (Information Content) = AB2H bits of information
where B = pc, or B = p0-p1 depending on whether the signal is a low-pass or a band-
pass function.
The information content of the signal part of the filter is not altered by the
presence of noise in the input because, for the type of filtering considered here,
the filter is constructed from a noise-free signal. The noise-rejection part of
the filter is dependent on the types of noise present in the input, and its effect
on the IC of the filter will be considered near the end of the next section.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-1-6
Further, the estimate of the IC of the signal was based on a discreet
system analysis. It could equally well be based on a continuous system analysis
with an integral replacing the sum in the computation of H.
3.0 Analysis of the Space-Bandwidth Product and the Information Content ot Spatial
Filters
In this section we will consider the space-bandwidth product and the information
content of spatial filters. To get precise answers to this problem we must develop
some fundamental parameters of the problem. In particular, we are interested in
detecting arbitrary signals in photo-records. We assume that the signals are known
and that the noise is additive, isotropic and homogeneous (See Reference 1). Under
these conditions it is appropriate to operate on the input data t(x,y) with a filter
whose transfer function is:
k S*(p,q)
H(p,q) N(p,(1) (3.1)
where S*(p,q) is the conjugate of the Fourier transform of the
signal of interest s(x,y)
.N(p,q) is the Fourier transform of the autocorrelation
function of the noise n(x,y)
is a constant.
These functions are defined by the following relationships:
S(p,q) = ITs(x,y) ej(Px clY) dx dy
and:
N(p,q) = olf?R (x y) e j(Px clY)dx dy
n
R (x,y)= n(u,v)n*(u+x, v+y) du dv
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-1-7
4 In general H(p,q) is a complex-valued function and is difficult to realize
in practice. However, the two-beam method allows H(p,q) to be recorded on ordinary
energy detectors. The method is based on the fact that a complex-valued function
can be recorded as a real valued function on a carrier frequency. It will be seen
later that the price paid for this method of realizing H(p,q) is the need for more
SBP in the filter function than the SBP present in the signal.
The most elegant method for realizing H(p,q) is probably the use of a modified
Mach-Zehnder interferometer as mentioned in Reference 1. An equivalent technique
and one better suited to the problem from a practical viewpoint is the modified
Rayleigh interferometer diagrammed in Figure 3.1. It is convenient to divide the
optimum filter function represented by Equation 3.1 into two parts - a part due
to the noise, and a part due to the signal:
H(p,q) =
N(p,q)
? k2 S*(p,q) (3.2)
The noise part of the filter function ki/N(p,q) is a non-negative function because
it is the Fourier transform of an autocorrelation function. Thus, it can be recorded
on a suitable medium by normal techniques and does not require the use of interferometric
techniques. The signal part of the filter k2 S*(p,q) is complex-valued if s(x,y) is
asymmetric, and interferometric techniques are necessary to record it on energy-
sensitive devices. The two parts of the filter can be made separately and placed
in contact after they are recorded to produce H(p,q).
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
POINT SOURCE OF
MONOCHROMATIC LIGHT
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
COLLIMATOR LENS Lc
SIGNAL s(x,y)
IN PLANE P1
TRANSFORM LENS Lo
P2
REFERENCE LENS Lr
REFERENCE POINT SOURCE
FIGURE 3.1
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-1-9
The signal whose filter we desire is placed in one-half of the light beam
in plane P1 of the system shown in Figure 3.1. A small lens called the reference
lens is placed in the other half of the beam so that it produces a point focus of
light in plane 131, a distance b from the center of the signal in the x-direction.
For purposes of this discussion, the light distribution in the point focus can be
represented by a delta function of strength Ro. If the center of the signal defines
the x = y = 0 point of plane Pl? then the total light distribution in that plane is:
g(x,y) = s(x,y) + R06 (x-b,y) (3.3)
Lens Lo takes the two-dimensional transform of g(x,y) and displays it in
plane P2 as G(p,q) where:
G(p,q) = S (10,0 Ro elPb
(3.4)
The recording medium used in plane P2 is always an energy detector that senses
intensity or G(p,q)
12.
It is not necessary that the detector be silver photographic
material, but we will use it for sake of illustration. A detector can also be used
that produces a phase variation rather than a transmission variation without changing
the results of this analysis. The photographic material is exposed and developed
so that its specular amplitude transmission is proportional to the exposure or:
Ta(p,q) = a - bE(p,q) (3.5)
where a and b are photographic constants as diagrammed in Figure 3.2.
By appropriate choice of Ro and the exposure, we can insure that E(p,q) lies
on the linear portion of the curve between El and E2. Since the exposure is proportional
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
0
Ei
EXPOSURE
FIGURE 3.2
E2
E(p,q)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-1-10
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
to IG(p,q)12, the transmission of the film is:
Ta(p,q) = a - b I G(p,q)I2
where G(p,q) is given by equation 3.4.
Substituting equation 3.4 in equation 3.6 and expanding, we have
(3.6)
- 2, -jpb
Ta(p,q) = a - b Ro +I S(p,q)I2 + Ro e S(p,q) + Ro eiPb S*(p,q)
A-1-11
(3.7)
We note that the last term in equation 3.7 is proportional to the second part of
the optimum filter function as represented by equation 3.2. It is, however,
multiplied by a linear phase factor, and is comingled with other terms. The optical
system that is used to process the data f(x,y) uses this linear phase factor to
automatically separate the term of interest from the others. This optical system
is shown in Figure 3.3. The data to be processed f(x,y) is placed in plane Pl.
Lens Lo displays its Fourier transform F(p,q) where F(p,q) is modified by Ta(p,q). Lens
Li takes the Fourier transform of this modified distribution and displays it in plane P3
where:
0(-x,-y)7 F(p,q) Ta(p,q) el(px + qy)
dp dq (3.8)
By
using the convolution theorem, equation 3.8 can be rewritten as:
0(-x,-y) =
ff(u,v) ta (u+x, v+y) du dv (3.9)
The negative signs associated with x and y result from the fact that successive
transforms in optical systems always introduce a positive kernel function. In
the following discussion for convenience we will drop the practice of noting this
rotation of the image through It radians.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
POINT SOURCE OF
MONOCHROMATIC LIGHT
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Lc
OBJECT
Lc,
SPATIAL
FILTER
P2 L1
f(x,y)
FIGURE 3.3
RECOGNITION
PLANE
P3
F(plci)Ta(plq) 0(xly)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-1-13
4 Our primary task at this point is to determine what minimum value of b
will allow the system to operate successfully. For this purpose it is easier
to use the representation of the output 0(x,y) as given in equation 3.9. The
first step is to find the impulse response of the total filter function. This
is done by removing f(x,y) from the system, and inserting a lens that will produce
a point source of light on the optical axis in plane Pl. The light distribution
in plane P. will then be the Fourier transform of Ta(p,q). In this analysis we
assume unity magnification for convenience, but the results are unaffected by
this assumption.
The Fourier transform of Ta(p,q) is:
ta (x,y) = 42 Ta(p,q)
(px + qy) dp dq
and by making use of equation 3.7 we have:
ta(x,y) = 1
41t2
43-t2
If
-bR02)1
(px + qy) dp dq
if I S (p, 012 el (Px +
clY)
dp dq
dp
+ qy]
1-31Q---L/TS(p,q)
dp dq
4n2
bRo
jO s*(p,q) e
r j[P(x+b)
42 dp dq
(3.10)
The first term of equation 3.10 is a 6-function of magnitude a-bRo2 located at
x = y = O. The second term is the autocorrelation function of s(x,y) and is
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
centered on the optical axis:
431.2
A-1-14
1 S I (p,q)12 ej(Px 00
dp dq = b infs(u,v) s*(u+x, v+y) du dv (3.11)
If the signal has lengthlx in the x-direction, then the right hand side of equation
3.11 has length 2Zx in the x-direction.
The third term of equation 3.10 is bRo s(x-b,y) which is the signal displaced
from the optical axis by a distance b. Similarly, the fourth term of equation
3.10 is bRo s*(-x-b,-y) which is the conjugate of the signal with reversed coordinates
displaced from the optical axis by a distance -b. Each of the latter two distributions
have length& in the x-direction. The impulse response of the total filter function
is illustrated in Figure 3.4. Thus, ta(x,y) is given by:
t(x,y) = (a-bR02) 15(x,y) -bs(x,y) * s*(-x,-y)
- bRo s(x-b,y) - bRos*(-x-b,-y) (3.12)
From equation 3.9 we see that if we remove the impulse function from the system
and replace it by f(x,y), the output of the system is given by:
0(x,y) = f(x,y) * ta(x,y) (3.13)
where * denotes convolution. (Note that * also denotes complex
conjugate, but it should be clear from the usage which operation
is intended).
Using equation 3.12 in equation 3.13 yields:
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
POINT SOURCE OF
MONOCHROMATIC LIGHT
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
P1
Lo
FIGURE 3.4
P2
Ta(p,q)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
P3
.1s(x?bly)
(a?bRP)8(x,y)+
s(x,y)*s*(?x??y!
s*(?x-1),?y)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
0(x,y) = (a-bR0
2 ) f(x,y) * 6 (x,y)
- bf(x,Y)* s(x,Y) * s*(-x,-y) - bRof(x,y)*s(x-b,y)
- bRo f(x,y) * s*(-x-b,-y) (3.14)
A-1-16
The first term of equation 3.14 has length Lx in the x-direction where Lx is the
length of f(x,y). The second term has length Lx + 42x. The third and fourth
terms each have length Lx + Zgx, but they are centered at x = +b. The minimum
value of b is chosen so that the various terms do not overlap. Both the first
and second terms of equation 3.14 are centered on the optical axis, but since the
second term occupies more space in the x-direction it determines the minimum length
of b. Thus, we have the condition for non-overlap as:
or,
Ibl 5 1
2 Lx + 4/x + 2/x
2 x
central order one side order
bl 5 Lx + 3ex
(3.15)
The minimum value of b occurs when the inequality holds, thus b = Lx + 3.
This has been a somewhat lengthy analysis merely to determine the minimum
value of b, but it is needed to gain an accurate picture of the SBP and the IC
of the spatial filter. It has also served as a review of the fundamental concepts
and gives some insight into the operation of the system.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-1-17
We are finally in a position to determine the SBP and the IC of the filter.
The optical system used to construct the filter was shown in Figure 3.1. It is
shown again in slightly modified form in Figure 3.5. To determine the SBP and
the IC of the filter we must find the value of four parameters. These parameters
are the area A of the filter, the highest frequency pc) that it contains, the
bandwidth B of the filter, and the number of states that each resolution element must
be able to assume.
The area of the filter is determined by the maximum frequency contained
in the signal. We regard the filter as a limiting aperture of a lens (effectively
the aperture of lens Li in Figure 3.3). The application of Rayleigh's criterion for
resolution shows that the filter must have a radius R = XD po, where X is the
2
wavelength of the illumination, D is the distance from plane P1 to plane P2 in
Figure 3.5, and pc) is the highest frequency contained in the signal. Thus the
area of the filter is A =2711. x1 D2 p02.
The highest frequency that the filter must record is determined by the signal
length ix and the data length Lx, both lengths measured in the x-direction (or the
direction normal to the carrier frequency). From figure 3.5 we see that the maximum
distance from the reference point to the edge of the signal is b + 1/2 ix, but
b = Lx 32x so that the maximum distance is Lx + 7/2 ix. Referring to Figure 3.6,
we have two points separated by a distance Lx + 7/2 ix, lying in a plane a distance D
from the filter plane.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
POINT SOURCE OF
MONOCHROMATIC LIGHT
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Lc
L0
P1
P2
Lr
FIGURE 3.5
REFERENCE POINT FOCUS
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
FIGURE 3.6
(SEE FIG. 3.5 FOR DETAIL)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
= 0
A-1-19
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-1-20
Let us denote a ray from the reference source image to an arbitrary
point p by r1 and a ray from the extreme edge of the signal to the point p by
r2. The light distribution in plane P2 as a function of p due to these two points
whose amplitudes are normalized to unity for convenience, is:
f(p) = eikri eikr2
We are interested only in the magnitude of f so we have:
2
(p)
2 = lejkrl e21 = 2
1 cos k (r1- r2)] (3.16)
Writing r1 and r2 in terms of p and D, we have:
r1 =D
r2 = D
1 + 1/2(p 2 +
1 + 1/2
p-Lx - 7/2 2x)
+
(3.17)
We can truncate the binomial expansions after the second term because D >> p. Using
equation 3.17 in equation 3.16 gives:
If(p) r= 2 L 1 + cos
kD
2
(p/D)2 _( P - Lx - 7/22N
(3.18)
The frequency is obtained by differentiating the argument of the cosine function
with respect to p:
xmax
1 d
2A dP
1
kD
2A
kD
2
(p/D)2 - p-Lx - 7/22 2} ]
p/D2 p - Lx - 7/22x]
2
1 Lx + 7/22x1 (3.19)
Approved For Release 2005/05/02 CIA-RDP78604770A002300020010-5
XD
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-1-21
Similarly, the minimum frequency in the x-direction is found by using the reference
point and a point taken at the nearest edge of the signal. This frequency is given
by
1
K.
D Lx + 5//ex
(3.20)
The bandwidth in the x-direction is given by BK = pxmax - pxmin = Lx/Ok,D)
The bandwidth in the y-direction is By = Zy/(klOrPhere Ly is the signal length in
the y-direction.
Since we have computed the bandwidths in two orthogonal directions,
it is also customary to compute the area of the filter as though it were a square of
side 2R rather than a circle of radius R. The area of the filter then becomes
A = %2D2p02.
Finally, we must determine the number of states that each resolution element
must be able to assume. As stated in section 2, the information content of a
signal must be equal in every plane in an aberration-free system that has adequate
bandwidth. Recall that the IC of the signal was equal to Ap0
2 H bits, where A is
N
the area of the signal pc, is its maximum frequency, and H = -
IC of the signal is then given by:
ICs =xy p02 s
where the area of the signal is .8 x y.
pt log2 pi
i=1
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
The
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-1-22
The IC of the filter is given by ICf = A Bx By Hf or:
2 2 2
ICf D po (-12?%D) '?7)1if
= Lx 2y po2 Hf
(3.21)
By comparing the IC of the signal with equation 3.21 we see that we must have
= H. In the event that each state has equal probability of occurrence in
both the signal and filter, we have:
EI = log2 Nf = Hs = log2 Ns (3.22)
or that each resolution element in the filter must be able to assume exactly the
same number of states as each of the resolution elements in the signal.
The space bandwidth product of the filter is given by:
SBPf = px max px,yX2D2po2
2
= (Lx + 7/22x)2y pc)
whereas the space bandwidth product of the signal is:
9
SBPs =22p
x y o
(3.23)
(3.24)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-1-23
It can easily be seen that the SBPf must be much larger than the SBPs. This
is due to the fact that the filter is effectively a band-pass function while
the signal is a low-pass function. The filter contains a band which is "empty"
between zero frequency and (Lx + 5/2 ix)/(XD)i/mrn which is needed in order to
properly process the data. Thus it is seen that the IC's of the filter and
signal are equal whereas the SBP's are not. In many cases R2
where: R is the radius of the lens
f is its focal length
u,v are coordinates in the lens plane
k = 2TtA,
t is the thickness of the glass
Approved For Release t101)5/05hOkt CiA-RDIRSBOF4T-1114(t1123110020010-5
2.12)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-2-16
The solution when a practical liquid gate is used follows this same line
of analysis except that we have two such glass plates and a space filled with a
liquid of refractive index n2. In passing, we should note that the amount of
qpherical aberration introduced by the glass is independent of its axial position.
We are now ready to place requirements on the liquid gate itself. First,
every surface should be flat to at least X/4, with a 08 requirement preferable.
The critical tolerance on the gate is the permissible wedge angle. This is not
of importance if the gate is stationary, but if a scale search is carried out by
moving the gate axially in the system, then the wedge angle is important. The
reason is that the point focus created by the objective lens will shift as a
function of the axial position of the gate due to varying optical levers. The
situation, greatly exaggerated, is shown in Figure 2.7.
Suppose we wish to move the gate over a range as to carry out some given
percentage scale search. Denote the wedge angle of the gate by a and the distance
from the lens to the frequency plane by r. The deflection of the point focus when
the gate is very close to the lens is 6= r sin a. The deflection when the gate
is in the other extreme position is 61 = (r-s) sin a. For the system to perform
well, the difference between 6 and 61 should be less than 1/10th the diameter of
the Airy disk produced by the objective lens. The size of this disk is:
1. 22 X r
(2.13)
where R is the radius of the lens. Using this criterion gives:
6 - 6 < d/10
1 =
sina <
Approved For Release 2005/05/02 : CIORDP781304770A002300020010-5
1.22X r
1OR
1.22 r
(2.14)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-2-17
Equation (2.14) then sets the limit on the maximum amount of total wedge angle
that the gate may have when it is filled with liquid of the proper refractive
index.
The next element in the optical processor is the spatial filter. From a
performance viewpoint this element must not introduce excessive phase errors.
If photographic film is used, it must not introduce more than )44 wavelength
of aberration over its aperture. Film thickness variations must be reduced to
this level by immersion in a liquid of matching index or by using a thin emulsion
coated on glass plates of suitable flatness. If photoplastic film is used, the
combination of the substrate and coating must also satisfy these requirements. A
few comments on the required positional accuracy of the filter will be given at
the end of this section.
The next element in the system is the imaging lens. Since it is the only
element in the system that must operate over a large field, its design has to be
different from the other elements. First, its aperture must be adequate to receive
all the diffracted rays, unless some auxiliary elements are used to bring the
important diffracted bundle of rays parallel to the optical axis. The thickness
of the glass substrates for the record and spatial filter must be considered as possible
sources of aberration.
Since the imaging lens will be used axially along with the input liquid gate
and the output sensor, it need only be corrected for one pair of conjugates and
one wavelength of light. The resolution capability of the lens must be adequate to
image the input scene well over the entire field. The image plane should have as little
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-2-18
curvature as possible, or could be designed for a particular type of sensor that
might be curved. The demagnification of the lens should be made compatible with
the size of the input record and the sensitive area of the detector.
The final element in the processor is the output sensor. Its resolution
capability as a function of its size should be adequate to resolve the input record
at the required amount of demagnification. Its sensitivity should be such that it
can handle data at the chosen rate.
3.0 Analysis of the Filter Making Subsystem
The optimum modulation transfer functions for the elements in the filter
making subsystem are very similar to those placed on corresponding elements in the
processor. In fact, some of the elements can be identical to those used in the
processor. Whereas the configuration of the processor is fixed, the configuration
of the filter maker is not. However, a typical subsystem will contain the elements
shown in Figure 3.1.
The laser, condenser lens, aperture and collimator are subject to the same
requirements placed on these respective elements in the processor; in fact, they
can be duplicates to save on cost. However, the laser should have a fair
degree of temporal coherence as well as spatial coherence in order for this instrument,
which is basically an interferometer, to work well. Generally 500-600 mm of temporal
coherence is adequate.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
LASER
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
COLLIMATOR OBJECTIVE LIQUID GATE
NOISE ELIMINATOR
OBJECT
FILTER
RECORDING
PLANE
ATTENUATOR
NOISE ELIMINATOR
FIGURE 3.1
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-2-20
The objective lens is identical to the one used in the processor except
that it can be of smaller aperture if the signals to be detected are small
compared to the field of view used in the processor. In fact, it may not be
necessary to correct this lens for the spherical aberration introduced by the
liquid gate that contains the signal, particularly if the effective relative
aperture of this lens is f/50 or less. Hence, a simple single element aspheric
lens with transfer function similar to that given in equation 2.14, but with smaller
aperture, is adequate. Note, that because of the scale changing feature of the
processor, it is not necessary that the two objective lenses have matched focal
lengths.
The liquid gate has the same flatness requirement as the one used in the
processor, but the wedge angle is not critical. A small amount of wedge in this
gate merely changes the carrier frequency by an insignificant amount. Also,
because of the scale change feature, it is not necessary to match the thicknesses
of the two gates.
The reference lens can also be a single element aspheric because its
purpose is to create a diffraction-limited point focus of light. This point focus
should occur in such a plane that the optical paths from the frequency plane to
this plane and the signal plane are equal. An aperture used at the focus will reduce
the "noise" in the reference beam. We may also wish to place an attenuctor in the
reference beam to control its amplitude. This element should be placed near the
point focus but ahead of the aperture to minimize aberrations. The use of
attenuating media in the signal beam is not recommended.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-2-21
Finally, the maximum performance of the optical processor will not be
realized unless the photographic film is positioned properly in the filter
making subsystem. The offset frequency present on the filter acts much the
same as a thin prism if the light is monochromatic, and will introduce astigmatism
in the processor. This astigmatism can be reduced to a minimum by positioning
the film in the filter making subsystem such that it is normal to the bisector
of the angle between the signal and the reference point as seen from the film plane.
This technique is equivalent to using a thin prism at its minimum deviation angle.
4.0 Summary and Conclusions
In this memorandum we have analyzed the requirements to be placed on each
optical element in the recognition system and the filter making subsystem. All of
the lenses except one can be single element aspherics. The use of single element
lenses reduces reflections in the system to a minimum. All elements should be coated
with anti-reflectant coatings to further reduce reflections. All elements should be
made of highest quality glass to avoid bubbles and striae. Polishing operations
should be done with extra care to produce a blemish-free surface.
The care of the optical elements after construction is almost as important as
the design and construction phases. Fingerprints and smudges are irreversible
processes and cleaning with solvents is likely to leave residues. All components
should be fitted with lens caps when not in use to prevent contamination.
In conclusion, the optical systems described here are not sophisticated ones
in terms of the types of lenses used. The nature of the process requires, however,
that the individual elements be well designed and kept in good alignment and clean
throughout the operation of the system.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-2-22
Note on Position Accuracy of Spatial Filter
This problem is technically not one of determining optimum transfer functions
because the optimum position for the filter is obvious. The problem is one of great
practical interest, however, and will be briefly discussed here In practice it is
often desirable to rotate or change filters while the system is in operation. The
question becomes one of determining how accurately this must be done to prevent
degradation below a certain point.
This problem is not an easy one to answer and cannot be solved by simple
analysis except for the white noise case. The non-white noise case can be
formulated, but requires the aid of a computer as well as knowledge of the noise
statistics to get any meaningful results. Personnel at the
have performed such an analysis on its Air Force and Army contracts, and the results
are available in the classified literature. (References 2 and 3)
Although the white-noise case is of little interest in itself, it is
interesting when compared to the non-white noise results. We are essentially
interested in four different results; the lateral and longitudinal positioning
tolerances for both the white and non-white noise cases.
We first treat lateral positioning errors for the white noise case. The
final results is that:
SNRD=
?if ls(u,v) 12e4(4u + pqv)
du
dv
If Is(u,v) 12 du dv
(A-1)
where SNRD is the signal-to-noise degradation; i.e. it is the ratio of the
SNR when the filter is displaced from the ideal SNR. Also, s(u,v) is the
signal; (Lsp,Aq) are the amounts of displacement in the p and q directions
in the frequency plane and are measured in radians/mm.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-2-23
The next case is longitudinal positioning errors for the white noise case.
The result is:
SNRD = ffis(u'N)
ff s(u,v) I 2 du dv
2 (AT' u +Aqiv)
du dv
2oxo
where Apt = A Z p
ql27-2?'0 Az
X22
pZ is the displacement of the filter
(A-2)
xo , yo are the coordinates of the center of the signal in
the input
2 is the distance from the input to the frequency plane
It is clear from equation (A-2) that the SNR is a function of the location
of the signal as well as the displacement of the filter. This illustrates that
the frequency plane is the only plane in which a filter can be placed without the
need for scanning. It should be stated that the result represented in equation
(A-2) is based on the assumption that the displacement is not so large that S(p,q)
varies appreciably. In view of the small relative apertures of the signals typically
used, this assumption is valid.
We now treat the case where the noise spectral density is not uniform.
The result for lateral positioning errors is:
SNRD
f S (p-Ap) S*(p)Tn(p) dp
f N(p-Ap) I Tn(p)S*(p) 12 dp
f J S(p)I2 Tn(p) dp
f N(p) I Tn(p)S*(p) 12 dp
where S(p) is the signal spectrum
Ap is the lateral positioning error
N(p) is the noise spectral density
T(p) is the transmission of the noise rejection part of the filter
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
(A-3)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-2-24
Finally, we have the result tor longitudinal positioning errors for the non-
white noise case:
SNRD
f s (p-a0)s*(p) Tn(p) dp
f N(p, z) ITn(p) S*(p) I 2 dp
1s(P) 12 Tn(P) dp
I NWITn(P) s*(P) 12 dp
(A-4)
, 2nxo
where 64)
N(p, pz) is the noise spectral density in the plane of the filter.
Note, that we have written equations (A-3) and (A-4) in one-dimensional
notation for convenience. The two-dimensional versions can be written
by inspection.
Admittedly equations (A-3) and (A-4) are of limited usefulness unless
N(p,q) is known, and certain assumptions are made about the construction of Tn(p,q).
The work mentioned in References 2 and 3 arrives at results for a specific case.
References
1. J. Kreutzer, paper given at Fall Meeting of the Optical Society
of America, New York City, New York.
3. Same as above except Report No. 6318-10-P, December 1964.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-3-1
Approved For Release 2005/W6iTb&-DP781304770A002300020010-5
The Use of Phase ModulatinE, Media for
Recording Data and for Recording Spatial Filters
I. Introduction
In this memorandum we will consider the feasibility of using phase modulated
media in the optical processor to replace conventional photographic film. The most
important advantage to be gained is that certain phase modulating media have the
capability of operating in real time or near real time. Also, there should be a fair
increase in the amount of usable light, and no chemical processing treatments are
necessary to obtain an image. Possible disadvantages are the potential increase in
the noise introduced into the system and the fact that phase media tend to be band-
pass operators. The latter property is not a problem if the phase media are used
for recording spatial filters which are themselves bandpass functions.
Our major purpose in this memorandum is to lay the theoretical ground work for
using phase functions and to determine whether any unusual techniques are necessary
to fully use the potential of such functions. Under separate headings we will discuss
the use of phase functions in the input plane for recording the data to be processed
and its use in the frequency plane for recording filter functions.
2. Use of Phase Modulating Media in the Input Plane
The basic optical system under consideration is shown in Figure 2.1. Its
operation has been described elsewhere and will not be repeated here. Suffice it
to say that the data to be processed is normally placed in plane P1 and will be denoted
by f(x,y). It is this element of the optical system that we wish to consider here. In
conventional systems f(x,y) is the specular amplitude transmission of photographic film,
and its related in some (usually non-linear) way to a given property of the real world.
The optical system is designed to display the two dimensional Fourier transform of f(x,y)
in plane P2 as F(p,q).
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
COLLIMATOR LENS
POINT SOURCE OF
MONOCHROMATIC LIGHT 131
/ I
I
1
I
I--
I
1
1
1
OBJECTIVE LENS
IMAGING LENS
/
DATA TO BE PROCESSED
FIGURE 2.1
/
SPATIAL FILTER
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
P3
/
READOUT
A-3-3
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
There are two ways to approach the problem at hand. One is to inquire whether
it is possible to convert f(x,y) into a phase function g(x,y) such that the rest of
the system's operation remains unchanged. This approach is of interest if one has
great quantities of data recorded on photographic film and wishes to process a phase
modulated version of it. The conversion can be carried out by reliefing the film or
by changing it into a refractive index variation. The exact method for doing this
is not of interest here.
The second approach is to use the phase modulating media to record the same
property of the real world that the photographic film recorded. In this case we are
not particularly interested in whether the resulting phase function is identical to
f(x,y) because the system's operation can be predicated on the use of this process.
The same comment could equally well apply to the first approach if the rest of the
system's operation has not yet been specified.
We will confine ourselves to the second of these two approaches. Suppose the
property of the real world that we wish to record is f(x,y). If this function is
recorded on a phase modulating media, we have for the phase function:
g(x,y) = exp jc f(x,y)] (2.1)
where c is a constant. If the total phase variation is less than 21-(1% , where X
is the wavelength of light, we can expand equation (2.1) to obtain:
g(x,y) = 1 + jc f(x,y) + (2.2)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A0023000200,40354
We can drop the higher order terms in the expansion if the restriction is met.
The Fourier transform of g(x,y) is:
,(px + qy)
= 6 (p,q) + jc f (x,y) dx dy
.1,
(2.3)
We note that the DC term is It/2 radians out of phase with the Fourier transform
of f(x,y). This fact is normally unimportant in practice.
So far we have tacitly assumed that the phase modulating media has a perfect
frequency response function for all frequencies contained in f(x,y). Some such
media (e.g. thermoplastic and photoplastic processes) are unable to record very
low frequencies because they cannot maintain long-term depressions. A typical
response curve might look something like that shown in Figure 2.2. We have also
allowed for some fall off at higher frequencies. Thus, the Fourier transform
of f(x,y) will not contain frequencies below the lower frequency cutoff Po of the
media. In many cases the inability to record extremely low frequencies is not
serious, but if the lower cutoff is very much above one line/mm, the performance
of the system will be reduced somewhat. The main point of interest here, however,
is that the two terms in equation (2.3) do not overlap when bandpass media are used.
Consequently, a small blocking spot can be placed on the optical axis in the frequency
plane to prevent the 6 function from continuing to the output. The inverse Fourier
transform of G(p,q) then becomes:
gi(x,y) = jc fi(x,y) (2.4)
where f1(xy) is now considered to be a modified version of f(x,y) due to the
modulation transfer curve of the phase media.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A- 3- 5
FIGURE 2.2
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-3-6
Thus, it appears that no unusual techniques are required to use phase
modulating media in the input plane for recording the data to be processed.
Since the input experiences two Fourier transforms, the light amplitude
distribution in the output plane is proportional to f(x,y) except that it is
modified by the transfer function of the phase media.
If the phase medium are to be used in the input plane, it must satisfy
some other general requirements. If it is supported by a substrate, that
substrate must not introduce more than X/4 wavelength of phase error into the
system and of course the phase medium itself must also satisfy this requirement.
The questions of speed of handling, ability to store information for long periods
oi time, resolution capability, dynamic range, etc., must also be considered before
phase media can be substituted for photographic films. Some possible problems due
to over-modulation will be treated in Section 4.
The fact that some phase media are inherently bandpass operators does not
necessarily limit their usefulness for recording data. It is a question of how
low is the low cutoff. If it is of the order of the reciprocal aperture in i/mm,
the effect will be quite negligible; in fact, even if it is as high as 1/4 to
1/2 2/mm, the phase media may still have a great deal of usefulness in typical recognition
systems. This is so because the noise rejection part of the optimum filter for signal
detection normally attenuates these frequencies anyway. They contain very little
information and usually will not be missed.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5 A- 3-7
3.0 Use of Phase Media in the Frequency Plane
Much of what was said in the previous section about the general require-
ments of the phase media also applies here. From a theoretical viewpoint we
have an interesting variation in the problem, viz., the phase media will only
work in the frequency plane if the function to be recorded is a bandpass function.
(This is not to be confused with the fact that the phase medium itself is of a
bandpass nature.) To see that this is so, consider a low-pass frequency function
H(p,q) that might be a candidate filter function. The associated phase function
is:
G(p,q) = exp
jc H(p,q)
n'7-1 1 + jc H(p,q)
- The impulse response of this candidate filter is:
g(x,y) = 1 + jc H(p,q)
.(px + qy)
el dp dq
(3.1)
(3.2)
=& (x,y) + jc h(x,y) (3.3)
Note that g(x,y) is the sum of the desired impulse response h(x,y) and an out
of phase 6-function. Since only one Fourier transform is carried out in going
from the frequency plane to the image plane, we have no opportunity to separate
these two images as we did in the previous case. Even if the phase medium is
inherently bandpass, we cannot separate the effects of the two terms of equation
(3.3), even though the two terms themselves may be separated. The reason for this
is that the output of the system is the convolution of the input function and g(x,y)
i.e. if, the input function is denoted by f(x,y) and the output is denoted by r(x,y),
then:
r(x,y) = f(x,y) * [6 (x,y) + jc h(x,y)]
= f(x,y) + jcf(x,y) * h(x,y) (3.4)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5 A_? 3_8
It is the second term of equation (3.4) that contains the useful information.
The first term is added to it and can be considered to be noise. Hence, if
H(p,q) is a low-pass function, phase media in the frequency plane will not yield
the desired operation.
Fortunately the filters used in the recognition system are themselves band-
pass functions. The use of phase media is not only possible, but in some sense
is better suited for this application than for recording the input data because
it is a bandpass media. Recall that the signal part of the optimum filter is given
by:
H(p,q) = R2 4. I s(p,012 + R e-iPb S(p,q) + R eiPb s*(P,q)
The associated phase function is:
G(p,q) = exp [jc H(p,q)]
PJ1 + jc H(p,q)
jc iS(p,q) 12 Re-jpb
S(p,q) + Re iPb S*(p,q)]
Taking the impulse response of G(p,q) yields:
g(x,y) = ? (x,y) [1 + jc R2] jc s(x,y) * s*(-x.,-y)
+ jc R s(x-b,y) + jc R s*(-x-b,y)
Now we see the effect of the bandpass nature of the filter function. The
(3.5)
(3.6)
(3.7)
interesting parts of the filter are on a carrier frequency so that they are
automatically separated from the objectionable 6 function that occurs on the
optical axis. The effect of the 6 function is exhibited in the zero order image
term that is of no intexest
Approved For 'Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-3-9
Although the phase medium can be used to record the signal part of the
filter, it is not so clear how it can record the noise-rejection part i.e.
1/N(p,q). One way that it could be done is to use the required noise-rejection
function in the filter making system so that the signal spectrum is already
modified by 1/N(p,q) and the effect thus recorded on the filter. This normally
leads to difficult mechanical problems unless the filter maker is constructed
so that it has two frequency planes. Then the noise-rejection filter could be
placed in one frequency plane and the film in the other, but this also leads to
systems with mechanical difficulties.
An alternate approach is to expose the phase medium twice, once to record
the signal spectrum and once to introduce 1/N(p,q). One would then hope that the
second exposure reduced the modulation frequency by the proper amount and in the
proper places to get optimum overall performance.
Still another way is to use both photographic films and phase media in the
recognition system. The noise-rejection filter could be recorded on the photo-
graphic film and placed in contact with the phase medium. Since the noise-
rejection filter does not have to be changed very often, we can maintain whatever
dynamic modulating characteristics the phase media might have. This approach would
require some changes to the filter making subsystem and the optical processor to
obtain good results.
It appears that the problem of introducing the proper noise-rejection function
into the system when required, will be the main difficulty in using phase media in
the frequency plane for recording filters.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-3-10
4.0 Phase Modulation Requirements
In this section we will consider the effect that over-modulation might have
on each of the two processes. The effect of over-modulation in the input plane
can be seen from equation (2.2). If c f(x,y) is not small compared to unity
(meaning that the phase modulation is small compared to 27t/X ), we cannot truncate
the series expansion after the second term. The more complete expansion for g(x,y)
g(x,Y) = 1 + jc f(x,y) c2 f2(xtY).
2!
c3 f3 (x,y) c4 f4(x,y)
3! 4!
(4.1)
Thus, when the medium is over-modulated, terms appear which essentially add more
noise to the process. The signals are distorted and will not, in general, correlate
with the filter. Finally, the various terms are out of phase with each other and
will lead to unusual interference effects in the output. The solution to the
problem obviously is to insure that over-modulation does not occur.
The effects of over-modulating the phase medium in the frequency plane are
slightly more difficult to obtain but are of greater interest. The basic equation
that we will deal with here is (3.6). For convenience it will be rewritten as:
where
g(p,q) =
1 + jc fA(p,q) + B(p,q) cos
A(p,q) = R2 I S(p,q) I 2
B(p,q) = 2R I S(p,q) I
0(p,q) is the phase part of S(p,q)
4(p,q) - pb
. . . (4.2)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
It will be helpful in carrying out the expansion of equation (4.2) to simplify
even further by dropping the arguments of the functions. Then equation (4.2)
becomes:
{G = 1 + jc [A + B cos (0 -pb)]
c2
7?=
[ A + B cos (0 -pb) ]2
A-3-11
3
-i c---[A+Bcos (0 -pb) ]3
+ . . . ] (4.3)
3=
={ 1 + jc [A + B cos
(0 -0)]
c2
- -27 [ A2 2AB cos (0 -pb) + B2 cos2 (0 -pb)]
C3
6
+. .
[ A3 + 3AB2 cos (0 -pb) + 3A132 cos2 (0 -pb) + B3 cos3 (0 -0)]
Further expansion followed by a collecting of terms gives:
G = 1 + jc A - c2A2
2
_ j c3A3 _ j c3AB c2B22
6 4 4
+ (jc B-c2 AB - c3A2B -
j c3B3
j ) cos -pb)
2 8
( j c3AB2 c2B2
4 4 ?
) cos (2 a - 2pb)
(4.4)
c3B3
- cos (3d- 3pb) + . ? (4.5)
24
First we note the presence of the expected harmonic terms. These terms do not
overlap with the terms of interest and can be disregarded. They do, however,
contain some energy and contribute to the inefficiency of the process. In passing
we also note that the harmonic terms have both amplitude and phase distortion and
are unacceptable for any use.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-3-12
The first term of equation (4.5) is equivalent to the zero order term. The
distortion terms cause some spreading of the central image in the output plane.
For example the term c3AB2is actually c3 [ R2 + 1 S (p,q)1 4R2 1 S(p,q)I 2.
The length of the central image in the output is Lx + 8ix, where Lx is the length
of the input signal and /x is the length of the signal to be detected, both taken
in the x-direction. When over-modulation is not present the length of the central
image is Lx + 42x , as described in Information Content of' Spatial Filters. Thus,
the carrier frequency would have to be increased somewhat to counteract this
phenomenon. Further, the presence of the distortion terms tends to use up the
available dynamic range of the phase media.
It is the second term of equation (4.5) that is of greatest interest because
it contains the desired information. Writing it with the appropriate arguments
and in ascending powers of 'S(pA) 12 we have:
( _ jc _ c2 R2 _ j c3 R4) Is(p,01
( _ c2 R - j 2 c3 R2 - j )
8
+ (- c3 R) IS(p,q) I
1 S(I),q) 13
cos (0(p,q) - ph)
First note that no phase distortion is introduced by the over-modulation. However,
two of the three terms contain amplitude distortion. This distortion produces what
might be termed a small signal suppression effect. The exact opposite might appear
to be the case from an examination of the terms given above, but it must be remembered
that less of the usable portion of the dynamic range is available for the term in
1S(P,q) 1. It can then be seen that those regions of IS(p,q) I that have low
nmplitudes tend to be suppressed by the distortion terms.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
A-3-13
It has been shown that over-modulation of the phase medium in either the
input plane or the frequency plane must be avoided if possible. While it is
true that the amplitudes of the distortion terms are lower than the amplitudes
of the terms of interest, the distortion reduces the efficiency of the process
and increases the noise factor.
It should be pointed out that over-modulation of photographic films will
produce similar results, except that the various terms will not be out of phase,
so that this phenomena is not unique to phase modulating media. Whether the
effect is more or less severe in the phase method depends on the available dynamic
range. This, in turn, is dependent on the amount of noise inherent in the phase
media. These questions cannot be answered without some experimental results on
which to base such calculations.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Section A-4
Obtainable Signal-to-Noise Ratios
A-4-1
1. Introduction
The ultimate factor in determining how well an automatic recognition system
operates is the signal-to-noise ratio (SNR) at the output of the system. A threshold
or decision level cannot be set and automatic detection is not possible unless the
SNR is sufficiently large. In this memorandum we will discuss the SNR that can be
expected from an optical recognition system. The problem will first be solved in its
most general form and then specific (although reasonable) assumptions will be made in
order to provide a few figures.
2. Mathematical Background
The input data f(x,y) is considered to be the sum of a signal of interest s(x,y)
and a noise background n(x,y). The noise background is said to be homogeneous if any
subportion of the data has approximately the same statistics as any other subportion.
The noise is said to be isotropic if the statistics show no preferred orientation.
The property of homogeniety guarantees that any particular image will have the
same noise statistics as any other image from the same general area. As a refinement,
one could divide the backgrounds into a subset of backgrounds - natural terrain,
urban areas, thermal backgrounds of IR images, etc. - to obtain somewhat more accurate
results. In this case the noise is assumed to be homogeneous within any particular
subset.
The signal is a Fourier transformable function of two space coordinates. Its
Fourier transform is given by:
S(p,q) = jr
-CO
s(x,y) e j(Px +
dx dy
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
(2.1)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
The noise has an autocorrelation function Rn(x,y) defined by:
R (x,y) = Lim 1
Ly.co A
Lx --)co
n(u,v) n*(u + x,v + y) du dv (2.2)
A-4-2
where A = 4 Lx Zy. In view of the assumption of homogeneity, it does not
matter whether n(u,v) in equation (2.2) is a complete subset of noise backgrounds or
any element selected from the subset. The noise can also be characterized by a
noise spectral density N(p,q) defined by:
00
N(p,q) = jrf Rn(x,y) ei(Px clY) dx dy
-oo
(2. 3)
It is desired to find the optimum filter function H(p,q) that will maximize
the ratio of peak signal energy to mean square noise energy. The signal part of
the output is:
rs(x,y) =
471[2
..(px + qy)
S(p,q) H(p,q) eJ dp dq
and the mean square noise (MSN) is:
00
MSN = 1 N(p,q) I H(p,q) 12 dp dq
411 -m
(2.4)
(2.5)
We want to find the H(p,q) that will maximize the ratio I r5(0,0) 12 IM,
where the peak signal is evaluated at x = y = o. The peak value of the signal
is not dependent on its location so we can set x = y = 0 for convenience. From
equations (2.4) and (2.5) we have:
co
r5(0,0) 2 1 r
fS(p,q) H(p,q) dp dq
MSN
12
liTN(p,q) 1 H(p,q) 12 dp dq
-03
(2.6)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-4-3
Since N(p,q) is the Fourier transform of an autocorrelation function, it is non-
negative and can be used as a weighting function in the Schwarz inequality.
Applying this inequality to equation (2.6) we have:
Irs00,0 2
)1 1
MSM
00
with equality if, and only if:
2
IS(p,q)1 dp dq
N(p,q)
k S* (p,q)
H(p,q)
N(p,q)
where k is some constant. If H(p,q) satisfies equation (2.8), the SNR is:
co
SNR= 1
zot2
IS(p,q)I 2
N(p,q)
(2.7)
(2.8)
dp dq (2.9)
Equation (2.9) gives the most general result for the obtainable SNR. The
SNR can be calculated as soon as S(p,q) and N(p,q) are known. Unfortunately,
the calculation of SNR cannot be done by hand in most practical cases. Before
passing on to a specific problem that can be calculated by hand, it should be
noted that if N(p,q) is constant (white noise case), then the SNR is dependent
only on the total signal energy - and not dependent on the signal shape. This
fact has some relevance in determining what signal shape to use in the calculations
of the next section.
3. Specific Examples
The general formulation for the SNR was given in Equation (2.9). In this
section we will assume a specific form for both the signal and the noise spectral
density that are not too far from reality. First, the fact that the noise is
isotropic implies that N(p,q) has rotational symmetry. We will replace N(p,q) by
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-4-4
by N(p ) where p 2 = p2 + q2. The noise spectral density can be quite closely
approximated in some cases by a function of the form:
N (p ) = ci
1
(3.1)
where c1 is the average noise energy n2, and a is a constant that determines the
shape of N(p ).
A somewhat better approximation to N(p ) might be attained by using a
function of the form:
N(p ) = 01
1
1 + a Ip1+ a2 p 2 + a3 1p13 +
(3.2)
where al, a2, . . . an are chosen to fit the particular N(p ) that is present.
We will use Equation (3.1) in the calculations of this section, although the
use of Equation (3.2) would not change the basic methods involved.
To simplify the calculations, we will choose a signal shape that also
has rotational symmetry so that we can replace S(p,q) by S(P ). Then Equation
(2.9) becomes:
SNR =
21(
1
42
ff0
1
N(p)
Is(P ) 12
N( p )
dp dO
p dp (3.3)
The choice of signal shapes is, at best, a compromise. We must have a signal
shape that will lead to conservative estimates of the SNR, but will, at the
same time, allow us to compute the SNR in closed form. A signal form that
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
satisfies both these requirements is:
_b(x2 .1_ y2)
s(x,y) = c2 e (3.4)
where c2 is the peak amplitude of the signal and b is a constant. The
Fourier transform of s(x,y) is:
co
jre-b(x2 Y2) ei(Px clY) dx dy
S(p,q) =
=4c2
=4c2
-bx2
cos px dx
e-by2 cos qy dy
e-p2/4b) ( 1 Tr e -(12/4b)
b
Substituting p2 = p2 + q2, we have:
c2
2/4b
S( p) = e P
Substituting (3.5) and (3.1) into (3.3) gives:
SNR
21t
2/4b 2
1;51 e-
1
c 1
l
+ /a)2
2
C2 Tr
2b2 n2
c22
2n02
pd p
( p p 3/112) e- P 2 /2b dp
b+
2b2
a2
2
a2
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
(3.5)
(3.6)
A-4-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-4-6
Equation (3.6) gives the expected SNR in terms of the peak signal amplitude in
the input, the average noise in the input, and the parameters a and b. Figure
3.1 gives a plot of (
for four values of the parameter a. Past experience
indicates that a should take on values around 1/2 for some real world backgrounds.
Thus, we use a = 1/8, 1/4, 1/2 and 1 for our plots of the noise spectral density.
2
In Figure 3.2 we have plotted ( 2no SNR
) as a function of b for the four
C22
values of a. These curves are asymptotic to l/b for small values of b and are
asymptotic to (2/a2) for large values of b.
To get some meaningful results, it is necessary to determine what values a
and b are likely to assume in an actual system. From some data we acquired some
time ago, it appears that a 1/8 for aerial photographs of natural terrain at a
2500:1 scale. The value of b can be determined from Equation (3.4) if we know the
length of the target (actually, theIliameter" of the target because we are working
with rotationally symmetric signals). Here we will use that value of b that will
cause s(x,y) to drop to its half value point at the edge of the signal (see Figure 3.3).
Any other assumption, say using the half power point will also be acceptable as long
Figure 3.3
13 -,..ussica? v\ 01S e
S
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
0.1
0.01
N(p)
0.00
0.000
0.0000
A- 4 -
Approved For Release 2005/ I
i :I-
::IA
,.\
10%.
_I
-
% % N
It N ?
II N
It N
I% ?
I'
I I
I 1
1
,
1 I
1
1.
1
1
?
1
?
1.
?
?
?
1
?
?
?
?
VI '
?
-S
?
?
?.
IIIIEIMIIIII
IIIIIIL%..
a = 1 2
?
e
111016.11011.....
?
?
?
ea; I/4
:
?
?
?
?
e
?a:1/8
?
?
1
2 3 4 5 6 7 13 9
p RAD./MM
10
II
Approved For Release 2005/05WPM-141478B04770A002300020010-5
1
1000
100
SNR(2q)
C22r
10
1.0
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
0.001
0.01
0.1
FIGURE 3.2
1.0
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
10
100
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5 A-4-9
as it is consistent. I choose to use the half value point because this results
in a half power point when we use the signal energy in the computations. The
value of b that satisfies this requirement is:
b = 4 (1.18.2
) = 2.8/D2
D
where D is the diameter of the target. In an aerial photograph at 2500:1 scale,
a target of 20 ft. diameter will have an image size of about 2.5 mm. Thus, for
that target b = 0.45. From Figure 3.2 we see that:
2
1 Tr
SNR = c- (128)
2n02
If the peak signal amplitude in the input is c2 = 0.5 and the average noise is no = 0.5,
then:
SNR = i?(128) 200
= + 23db
It is reasonable to assume that c2 no in the input so that the output SNR
will vary accordingly.
Before concluding this section we must examine the validity of equation (3.6)
for both large and small values of b. A large value of b implies that the signal
is very small and a small value of b implies that the signal is large. The result
expressed in equation (3.6) is valid at least up to the point where the signal fills
the system's aperture. The value of b can be obtained by requiring that the signal
as given by equation (3.4), be less than or equal to 0.223 at the edge of the aperture
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-4-10
so that we do not lose more than 5 percent of the signal's energy. If the system's
aperture is 2A, we have:
If 2A = 100 mm, then
2 ? (1.05)2
b A/2
.2-5:
b >
(1.05)2
2 (50)2
b > 0.0002
Thus, it is safe to use Figure 3.2 for any small value of b that is shown. The
validity of equation (3.6) for Large values of b can be estimated from equation
(3.5). Again, if S(p ) is less than or equal to 0.223 at the frequency plane
aperture cutoff, no more than 5 percent of the signal's energy will be lost.
Suppose the system is band limited at Po rad/mm. Then from equation (3.5) we have:
For po = 27t (50 /mm) we have:
fd2/4b (1.05)2
2
b
2(314)2
4(1.05)2
50,000
This value of b is well outside the range of values of b as shown in Figure 3.2.
Thus, the curves shown in Figure 3.2 can be used for all values of b as given.
4. Summary and Conclusions
In this memorandum we have derived a general expression for the SNR at the
output of an optical recognition system. The SNR can only be calculated in closed
form for certain signal shapes and for certain noise spectral densities. For the
more general case a computer is ususally necessary to compute the SNR. A computer
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-4-11
solution also has the advantage that the SNR as given in equation (2.9) can be
found as a function of finite limits so that the effect of the system's bandwidth
can be seen.
An estimate of the SNR for assumed signal and noise spectral density functions
was made in Section 3. The results indicate that the SNR can vary widely depending
on the exact form of the signal and noise functions. In practice we have found SNR
of about 15 db, so these estimates are fairly realistic.
A note of caution is in order, however, when using these results. We have
assumed that the entire system is linear. The filter function, however, is
recorded on a physical medium that of ten is non-linear. Again, this refinement
of the analysis can only be made with the assistance of a computer.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Section A-5
Information Capacity of a Recognition System
A-5-1
1. Introduction
The problem is one of showing that the optical processor can handle large,
highly detailed images. We will also consider the limitations that may be introduced
by some of the system parameters.
This case is closely related to the material covered in "Information Content
of Spatial Filters", except we cannot properly consider the information capacity
of a noise-free system. We can, however, talk about the space-bandwidth product
(SBP) of such a system. The major problem is that the types of noise in an optical
can vary widely and cannot be very well defined. The noise may be due to stray
light, blemishes on the optical elements, or noise in the recording media or sensor.
The effects of noise have been described by G. Toraldo di Francia in Optica Acta,
Vol. 2, No. 1, April 1955.
2. Space Bandwidth Product of the Optical System
In the section on Information Content of Spatial Filters we considered the
SBP of the signal to be detected and the SBP required of the spatial filter. We
did not consider the SBP of the input data, but it can be obtained by a simple
extension of that analysis. Recall that the SBP of any signal is equal to the product
of the area of that signal and the highest frequency that it contains. We will assume
that the input data is a low-pass function, as is almost always the case.
At this time we are only interested in matching the SBP of the optical system
with the SBP of the input data. We will use the system shown in Figure 2.1 as the
basis for this analysis. The data to be processed f(x,y) is placed in plane Pl, a
distance ifrom the frequency plane P2. The Fourier transform of f(x,y) appears
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
OBJECTIVE LENS
INPUT DATA f(x,y)
Pp/P2
-
IMAGING LENS
FIGURE 2.1
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
OUTPUT
P3i
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
in plane P2 as:
xi _fk
F(? ,n) = g(? ,2 ) if ) e
(Ex + ny)
f f dx dy
A ? j (?2+n2)
where g(? , =
A-5-3
(2.1)
In this case c ,n are the actual distance variables in plane 132, k = 211/X , X is
the wavelength of light, f is the focal length of the objective lens, and A is the
amplitude of the light in the collimated region. The presence of the factor g(E )
is not of importance in this analysis and will be neglected. Equation (2.1) exhibits
the scale varying transform relationship; but since the input data has its smallest
effective scale and its Largest effective area when 2 = f, we will consider only that
case.
We will first consider what requirements must be placed on the size of the
frequency plane due to a small patch of the input data centered on the optical axis.
The size of the frequency plane is determined solely by the highest frequency in
this patch. Suppose the input data is band-limited to a frequency
the radius of the frequency plane must be:
1/2
2 2
R = (?0 +n ) = po X f
Po
2/mm, then
(2.2)
Thinking of it in other terms, suppose the frequency plane is limited in size to
radius Ro or more likely, the aperture of the imaging lens is the limiting radius,
then the maximum frequency that can be passed by the system is:
Ro
P
f
Lb= (2.3)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-5-4
Thus, the limit on how highly detailed the input data can be is ultimately
determined by the aperture of the frequency plane or the aperture of the imaging lens,
whichever is smaller. This is exactly what one might expect because, in the absence
of a filter, the resolution of the total system is limited by the imaging lens.
It is clear from a consideration of equation (2.1) that the location of this
small patch in the input plane has no effect on the required size of the frequency
plane. Again, in other terms, for a given bandwidth, the patch can be just as
highly, detailed regardless of its location in the input. Thus, it will not make
any difference if there are many such patches in the input plane. In fact, the
entire aperture of the system can be used to illuminate the signal. Hence, we
have shown that the optical system can process large, highly-detailed images without
sub-dividing them into smaller images for processing.
A good analogy to this situation can be made by considering an electronic
system. If the system has some given bandwidth, the duration of the signal is of
no importance. In general the bandwidth requirements are independent of the duration
of the signal, except for very short duration signals.
Perhaps the only question generated by this chain of reasoning is that it might appear
that the frequencies from the various subportions of the input will interfere and
produce some anomalous effect. This is not the case. The maximum frequency that
can be produced by these interference effects in the frequency plane is:
A
(2.4)
Any frequency so generated can be passed by the imaging lens provided it does not
vignette the field of view. Proper placement of stops in the lens can prevent this
occurring. Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5 A-5-5
Finally, there is an upper limit as to how highly detailed the input data
can be.
If the highest frequency po approaches 2/X, 2/mm, the diffracted waves
will be evanessent and will not propogate into the system. Very few recording
media have resolution capabilities that will produce such an effect, and no image
transmitted by a realizable lens can contain such frequencies. It should be
pointed out, however, that such frequencies can be obtained in holograms when the
field of view extends to very large angles.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Section A-6
Approved For Releaseann92iIRMEE8B04770A002300020010-5
A-0-1
1. Introduction
In this section we will discuss the performance of the optical recognition
system as a function of the actual image size. This material is closely related to
the section discussing the obtainable signal-to-noise ratio (SNR) in an optical system.
The reason for this is that the same measure of performance will be used, i.e., the
minimum required SNR will determine the minimum signal size. It differs from that
discussion in that we are now concerned with microscopic noise (film grain noise)
as well as macroscopic noise (the background terrain). One expects the background
terrain noise to enter this analysis in much the same way as before and expects the
film grain noise to further reduce the SNR.
The fundamental problem is to construct a good mathematical model of film
grain noise. We know, for example, that film grain noise is a multiplicative
process. This means that the amount of noise is dependent on the signal strength
and that the total input data must be described by a function f(x,y), where:
f(x,y) = s(x,y)ni(x,y) no(x,y)
In this representation no(x,y) is the background terrain noise, s(x,y) is the
signal, and ni(x,y) is the film grain noise. Since no(x,y) is considered to be
macroscopic in nature, we can assume that it is not affected to any great extent
by the film grain noise. As the signal size decreases we can expect the system's
performance to be quite highly dependent on the exact nature of the film grain noise
process.
As has been shown elsewhere (Reference 1), we cannot formulate the mathematical
theory for matched filtering when the noise process is multiplicative. The best we
can do is (1) make some approximations that reduce the problem to one of additive
noise or (2) use a computer to try to optimize the filter under the multiplicative
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
noise restriction. There is no guarantee that matched filtering is optimum,
of course, but the problem gets out of hand from the analytical viewpoint.
A-6-2
There is one feature about the types of signals usually detected in
recognition systems that allows us to pursue the former approach. The signals
tend to be unitormly illuminated. For example, if the signal is of the form
shown in Figure 1.1a, one could expect that the difference between the multiplicative
noise process and an assumed additive noise process would be significant. However,
if the signal is more like that shown in Figure 1.1b, the difference between the
multiplicative and additive cases should be small.
The way in which the many different methods of describing photographic film
noise enter into this problem is not at all clear. Researchers have discussed
the accutance, granularity, definition, resolution, etc., of photographic film, but
it is difficult to incorporate such quantities into the calculations of the SNR.
It appears that the best method of describing the film grain noise for the purposes
of this study is to say that the film grain noise spectral density is uniform for
all frequencies of interest. Different films will then be characterized by the
amount of noise that they introduce.
For the type of signal shown in Figure 1.1b, we will assume that we can write
the total input function as:
f(x,y) = c s(x,y) ni(x,y) + s(x,y) + ni(x,y) + no(x,y)
where c is some constant. Here we have divided the multiplicative noise process
into a sum of signal times noise and a sum of signal and noise. We will assume
that c is sufficiently small so that we can neglect the first term of f(x,y) as a
first order approximation to the true situation. Thus, if N,(p,q) is the spectral
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
(a)
A-6- 3
FIGURE 1.1
(b)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-6-4
density of the background terrain and Ni(p,q) is the spectral density of the film
:rain, the total noise process is N(p,q) = No(p,q) + Ni(p,q).
It is admitted that the approach taken here is less than what would be
desired. Under the circumstances it is about all that can be done, but it is hoped
1 hat this approach represents at least one step toward the real solution to the
droblem.
;!.. Mathematical Solution
We will draw on the results presented in the section "Obtainable Signal to
Noise Ratios" as a starting point tor this discussion. Recall that the SNR was
4i.ven as:
?,
r
PIS(P)12(1_4E_
SNR = ?I? i
2ff j N(p)
0
(2.1)
where both the signal and noise were assumed to have rotational symmetry. The
signal has the form:
with Fourier transform:
s(x,Y) = c2 e-b(x2 + y')
C 9
S(p) = `
2/
e- P 4b
The macroscopic noise spectral density due to the background terrain was
dssumed to be of the form:
No(p) =
9
no
I
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
(2.2)
(2.3)
(2.4)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-6-5
The microscopic noise spectral density Ni(p) due to film grain noise will be of
the form Ni(p) = n12. Thus, we are assuming that the microscopic noise has a
flat spectrum which agrees favorably with experimental observations. The total
noise spectral density is then:
N(p) =
no2
+ ni2
1 p2/a2
no2 4. n12 n12 p2/a2
1 + p2/a2
Substituting equations (2.3) and (2.5) into (2.1) gives:
SNR -
co -2/2b dp
c22rc (1 p2/a2) pe-p
- c22"
2b2
cZ2
0
2b2a2
no2 ni2 ^ n12 p2/a2
pe- P2/2b dp
2
+ nl 2 n12 P2/a2
no
_p2/2b
Ip
p3e
n02 + n12 + n12 P2/a2
(2.5)
(2.6)
(2.7)
The two integrals of equation (2.7) can be solved by letting p2 = u, 2pdp = du,
to get:
C,, It
SNR - --2--
4b2
e-u/2b
no2 + n12 ^ n12 u/a2
du
^ c22't r11 du (2.8)
4b2a2 no2 + :1:2/2:n2 u/a2
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010-k_6_ 6
These integrals can be put into standard form by letting n12 u/a2 = t. Then
we have:
SNR
2 2
c2 a
4b2n12
ta
e .1.317;17?
no2 4. n12 4_ t
dt
te2
co
t e 2bni2
c211 a2
4b2 n14 n02 n12 + t
dt (2.9)
From the "Tables of Integral Transforms", Vol. 1, Edited by A. Erdelyi, McGraw-
Hill, 1954, pg. 137, we have the following integral transform:
,X)
[ tv(t +0-1 e-Pt dt =r + 1) av e aPr (-v,ap)
0
Thus we can solve the two integrals of equation (2.9) by letting v = 0 and
v = 1 respectively. Further, we let p -al,_ and a= no2 4. ni2. Then
9bni.
equation (2.9) becomes:
SNR
2 2
C2 ?. r(i) e
4b2 nl2
cfPr (0,ap)1
(2.10)
We also have the following
where:
c22 It a2
(2) aeaP r aP)1
for the incomplete gamma functions:
= 1
El (ap)
= -EI (ap) + a-ap
(2.11)
(2.12)
(2.13)
(2.14)
(2.15)
4b2n12
relationships
r(1) = r (2)
F(Opp) =
r(-1,ap)
El(x) =Jr
0
a p
e-t dt
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-6-7
(2.16)
(2.17)
Using (2.12) - (2.14) in (2.11) we have:
2ga 2
SNR = c2 e aP El( UP)
4b2ni2
c22Trn2
4b2ni4
c2'ra2
4b2n12
aeoP e-cp
cp
(i')-ae E (ap)
aP (1 -can12)E1(ap) +
Using the values given previously foraand p, we have:
SNR
2 2
c2 Tra
2 2
4b n1
a2(n04n12) 1-
no2+1112
2 El
2bn12 nl
- a2(no4n12)-
- 2bn12
2b
;72?
Note that the first term of equation (2.17) is zero if n02 = 0, i.e. if the
macroscopic noise is zero. Thus we can also obtain the "white noise" solution
from equation (2.17).
Equation (2.17) is plotted as curve A in Figure 2.1 for the following
values of the parameters:
no2 = 0.29
ni2 = 0.01
a = 0.1
c22 = 0.30
We note that curve A is asymptotic to c22/2b2n12 for large values of b. The
asymptotic behavior for small values of b cannot be found because E1(x) is not
tabulated for x < 0.01. It appears that the SNR is growing very slowly as would
be expected.
In any event, this region is not of particular interest because it
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
SNR
iT
100
10
1.0
0.1
0.01
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A
0.1
1.0
FIGURE 2.1
10
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
100
1000
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5 A-6-9
represents large signals. We are interested in the behavior of the SNR for large
values of b, corresponding to small signals.
Suppose we require a SNR of + 3db for adequate detection of the signals.
From curve A we see that b must be less than or equal to approximately 23 for
the given values of the parameters. This value of b gives a lower limit to the
signal size (to the half value points) of 0.35 mm.
The number of parameters is too large to give a comprehensive family of
curves of SNR. We will, however, plot two other curves to get some idea of
2 + n
how a change in these parameters changes the SNR. First, suppose no 12
n12
increases, i.e. we have a film with less grain noise. In curve B of Figure 2.1,
_
we have n02 ? 0.297, n12 = 0.003, a = 0.1 and c22 = 0.30. The overall effect is
to move the curve to the right of the first curve without affecting its shape.
As we would expect, the maximum value of b increases for a given SNR which
shows that the image size can be smaller when fine grain films are used. The
opposite effect will be experienced for coarse grain films.
Next we observe the effect of the change of the parameter a i.e. a
different form of the macroscopic noise. In curve C of Figure 2.1 we have
no2 = 0.29, n12 = 0.01, c22 = 0.30 and a = 1/7 i.e. the same parameters as for
curve A except that we have increased the parameter a. The curve for the SNR is
not affected for large values of b (the film grain noise is predominant here), but
is somewhat lower for small values of b. This denotes reduced detectability of
large signals. This could be anticipated because less signal energy is received
in the output plane. Also, the opposite effect is expected for a decrease in the
value of the parameter a.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
A-6-10
3. Summary and Conclusion
In summary, we have attempted to find the minimum signal size that can
be detected when one considers film grain noise as the limiting factor.
In order to make an analytical treatment of the problem it is necessary to
make some approximations and assumptions with respect to the nature of the
noise process. Although the film noise is generally thought of as a multiplicative
process, it can be approximated by an additive process if the signals to be detected
are nearly uniformly illuminated. Under these conditions an equation was derived
for the signal-to-noise ratio. In using this equation it is first necessary to
determine the value of the average microscopic noise power n02 and the value of
the parameter a that controls its shape. These two parameters are not affected,
in general, by the particular film used. It is then necessary to determine the
value of n12 for the particular film used. The value of n12 is not independent
of the signal amplitude, but it varies in accordance with a curve similar to that
shown in Figure 3.1. Thus, given a value of c22, one can determine the proper value
of n12 for the type of film to be used. Equation (2.17) can then be plotted as a
function of the signal size as reflected in the parameter b.
In spite of the assumptions that were made, equation (2.17) exhibits all
the properties that one would expect from such a noise process and no unusual
results are in evidence.
25X1
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-6-11
4
FIGURE 3.1
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-6-12
Section A
Conclusions
t. There should be no difficulty in obtaining the necessary information content
in the spatial filter. The highest resolution requirement is that of the carrier
frequency which is the interference fringe pattern produced in the two-beam filter.
This usually amounts to 125 to 250 lines per millimeter which can easily be recorded
on any of several fine-grained emulsions.
2. The optical requirements of the processing device involve no impossible
conditions. The necessary optical quality is easily obtained and results in a
modulation transfer function that appears adequate for any purpose that can be
envisioned at present.
3. Phase-modulating material such as
)hotoplastic film can be
used to make spatial filters which perform in a manner very similar to amplitude-
modulating silver photographic filters. Mathematically, the two are interchangeable.
4. To calculate the theoretical signal-to-noise ratio it is necessary to assume
a grain size for the film and a noise level due to extraneous objects. Making
reasonable assumptions for aerial film the conclusion is that a signal-to-noise
ratio of 15 db should be obtainable. Translated into optical terms this is adequate
for automatic detection and agrees with experimental findings.
5. It has been shown that the upper limit of detail in the photograph which the
process can handle is 2000 lines per millimeter. At present there are few lenses or
emulsions in common use which will realize as much as 200 lines per millimeter in the
finished photograph, therefore the principle is adequate for a purpose considered in
this report.
6. The minimum size of detectable objects in an aerial photograph is about .3 mm
or .012 inch. This calculation was based on assumptions for typical grain size and
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A-6-13
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
noise. For fine-grained film the minimum size would be less, and for noise-free
subjects such microfilmed printed copy the minimum size is much less.
7. The general conclusion we have drawn from this theoretical study is that the
process is capable of much more than is needed at present, the real problems are
materials and techniques.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Section B-Equipment
B-1-1
The purpose of these tests was to settle several equipment design and
material problems which appeared serious in 1964. These covered such questions
as the effects of irregular glass or plastic emulsion support on the optical
system and the relative merits of glass and film spatial filters.
The individual tasks were:
B-1 The Effect of Poor Quality Glass Plates in a Recognition System.
B-2 The Effect of Aberrations due to Glass Plates in a Recognition System.
B-3 Effect of Vibration on the 20-Foot Optical Bench.
B-4 Means for Attenuating the Reference Beam.
B-5 A New Means for Attenuating or Intensifying the Reference Beam.
B-6 Relative Merits of Glass and Film for Spatial Filters.
B-7 Use of Liquid Gates.
B-8 Test of S.O. 243 Film for Spatial Filters.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
B-1-2
Section B-1
Effect of Poor Quality
Glass Plates in Recognition System
Definition
Poor quality in a flat glass plate is defined as any factor that may cause the
plate to show wavefront deformations in a beam of light. These may be surface - quality
defects such as lack of flatness of the faces resulting in variations in thickness of
the glass, and hence variations in phase retardation of a coherent plane wave or they
may be material quality defects such as variation in refractive index. In thin glass
plates the surface defects predominate.
atement of the Problem
Glass plates can be introduced in the "perfect" system of the recognition apparatus
in the following ways:
1. Use of a glass plate transparency .as an object to make the spatial filter.
2. Use of a glass plate transparency of the objects to be recognized.
3. Use of a spatial filter on a glass plate.
1. Glass plate transparency of an object. In general, the objects from which the
inters are made are small. The quality of most plates will not vary much over this
:mall area so that this is rarely a problem. If severe, the effect will produce a
poor spatial filter. If the object is a simple one resulting in well defined diffraction
.pectra, microscope examination of such a filter will show distorted spectra. In the
case of complex objects distortion is difficult to recognize.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5 13- 1-3
So far, in our experience we have had only one object on a glass support which
had defects serious enough to cause trouble. This piece of glass was unusually bad
and should not have passed inspection as photographic plate material. The condition of
the surfaces is difficult to describe except that they were "rough", a condition rarely
seen in fire-polished, rolled or polished glass. It appears that 99 percent of any
sort of glass photographic plates will be adequate for objects up to 10 mm size.
2. Glass plate transparency of objects to be recognized. There is considerable variatior
in the surface-quality of the glass in any box of ordinary plates. These plates are made
of "rolled" glass which tends to have a wavy surface in the direction of rolling and these
waves produce astigmatism. The plates may also be wavy crosswise and these two effects
can combine to produce:
a. Areas in which the surface results from two crossed cylinders. If
both cylinders have the same sign the surface is approximately spherical.
b. Areas in which the surface results from two crossed cylinders of different
sign. This produces a lens positive in one meridian and negative in the
other and is known as "saddle-shaped" astigmatism.
c. Areas in which the crossed cylinders are skewed and random in sign.
This is known as an "irregular" surface.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
B-1-4
When an appreciable amount of any of these aberrations is added to the
diffracted beam the diffraction spots will not fall on the proper areas of the
Aiatial filter and poor recognition will result. In small transparencies, 1 inch
quare and less, the defects due to using commercial plates appear negligible, except
In the case of an unusually bad piece of glass. For 4 by 5 inch transparencies the defect
in commercial plates seriously degrade the recognition. This case was examined in detail
,ind is reported under experimental tests.
.1. Use of spatial filter on a glass plate. The equipment we have used produces a
spatial filter not over 26 mm in diameter with most of the information in an area
1bout 5 mm in diameter. It appears that the small area allows the use of commercial
plates or film with good results.
'3cope of the Investigation
In experimental work such as described in this report it is convenient to use glass
photographic plates for transparencies. Glass is usually a more uniform substrate than
itlm, glass plates are generally coated more uniformly, they do not show linear coating
offects usually detectable in film and glass plates do not curl or shrink due to processing
or storage.
The purpose of this tests was to determine if the commercially available glass
photographic plates were satisfactory in the recognition system. These plates are
coated on selected rolled glass and can be expected to show the same optical defects
as window glass. If possible, it would be desirable to use them because they are easily
available in many types, low in cost and fairly easily cut to smaller sizes once the user has
mastered the art of cutting ass in total darkness. The two alternatives were the use of
microflat plates which are expensive, difficult to obtain and hard to cut, or the use
of liquid gates to
ARAligina08r WeAiittidegioit?ctii-Ftimb fetilgth,M002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
B- 1-5
Original Plan of Experiments
It is fairly clear that when wavefront deformations take place in the transparency
to be searched that some of the diffracred rays will be deviated from their normal paths
and will no longer fall on corresponding points of the spatial filter. The result
will be twofold, a loss of signal strength and a gain in noise or leakage. The total
result will be a reduced sin ratio and poor recognition. The original plan was to select
a typical simulated aerial photograph on a plate known to be nearly perfect, then using
a spatial filter for one of the objects to produce a recognition spot. The intensity
of the spot and its surroundings would then give the sin of a "perfect" plate. Then
additional clear glass plates containing known and measured defects would be added one
at a time and the sin measured for each case.
Several tests were made in this manner and useful results obtained, but two
practical difficulties made the work slow. Most of the plates contained appreciable
wedge angle which deviated the zero order and required repositioning the spatial filter
to compensate. It was difficult to align the filter with the badly aberrated image
produced by the worst plates. The increase in the "noise" component appeared in all
sorts of unexpected places due to the random location of local defects in the plates.
This made it necessary to search the entire picture area for increases in background
leakage. It was decided to postpone this part of the test until the T.V. readout
system was in use as this would largely automate the measurement of sin ratios. In the
meantime a simpler test was used which appeared reliable and correlated well with the
measurements already made.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
B-1-6
Revised Plan of Experiments
In the last paragraph the defects in the glass plate were found to deviate the
rays diffracted by the transparency resulting in a loss of signal and a gain in noise.
It had already been calculated that any deviations greater than half the diameter of
the diffraction disc size would be detrimental. The diffraction disc of the basic
system is determined by the entrance aperture, the aperture ratio of the collimator
lenses, their quality and alignment. It is about 20 microns in either of our
instruments. It clear glass plates were placed in the beam of the instrument their
defects would increase the size of the diffraction image and this could be measured.
Assuming the aberrated image to be uniform in intensity by neglecting "streamers", it
3hould be possible to estimate the area covered by this image and the resulting
Attenuation of the signal. This tests tells nothing about the "noise" component except
hat it increases. It does not seem practical to calculate the noise increase, but
since this is spread over a large area of the picture it is of relatively less
importance than the loss of signal in determining sin ratio.
Source of Glass Plates
Forty 4 by 5 inch glass plates, most of which had been used as transparencies
were selected. These plates were graded as to whether it had been possible to obtain good,
or or no recognition from them in previous tests. The emulsion was dissolved and the
plates carefully cleaned by a procedure designed to prevent etching or scratching the
J-ss. All of the plates were then tested in an interferometer adjusted to produce a
.,;ingle fringe. If the plate were perfect the view would remain a single fringe. Most
if the plates contained appreciable wedge angle which required readjusting the inter-
1_-ometer to the single fringe condition. If the plate contained defects the distortion
,t the wavefront was shown by the number and pattern of the fringes. This was recorded.
The two sets of results; the performance of the plate as a transparency and the test
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
of the glass support agreed in almost every case, acting as a double check on the
glass quality. The results fell into three categories, important for the present
purpose:
1. Transparencies which had shown good recognition all showed minor wavefront
deformations, usually less than one wavelength and this "regular" meaning
cylindrical or spherical.
2. Transparencies which had shown poor recognition showed one to two and a
half wavelengths of deformation, usually regular, or up to one wavelength
of irregular deformation.
3. Transparencies which had shown no recognition showed more than three
wavelengths of wavefront deformation.
Numerical Results
In the following Table the first column gives the wavefront deformation measured
in the interferometer. The second column gives the area of the aberrated diffraction
disc produced by this plate compared to the normal disc area. On the basis of simple
theory and the assumption of uniform illumination, the intensity of the recognition
spot should vary inversely with this area. Actually the aberrated image is most intense
at its center so that the recognition signal does not decrease this rapidly. The column
"Measured Signal" gives the results of the original plan of the experiment described
earlier. The values are arbitrary based on the intensity of the unaberrated signal
as 100 units. The column "Measured Noise" is also the result of the originally planned
experiment and gives arbitrary values, not related to the signal values, but based on
the intensity of the unaberrated noise as 1.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A00230002001(01-8
Measured
Wavefront
Deformation
Type of
Deformation
Relative
Area
Image
Relative
Intensity
Recognition
Signal
Relative
Noise
None
1
100
1
1/8
Regular
1
100
1
1/4
Regular
1
100
1
1/2
Regular
1
100
1
1
Regular
4
60
2
1 1/2
Cylindrical
7
40
3
2 1/2
Cylindrical
14
10
4
2 1/2
Irregular
20
5
5
Astigmatic
32
3 1/2
Astigmatic
15
3/4
Cylindrical
6
1/2
Astigmatic
11
2 I/2
Irregular
14
5
Regular
32
3/4
Cylindrical
8
1 1/2
Irregular
24
3/4
Irregular
10
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Conclusions
1. The glass on which commercial photographic plates is coated contains
surface irregularities which result in wavefront deformations in the
recognition process which can reduce the s/n ratio of the signal.
2. For very small systems where the area of the plate is not over one
square inch, most plates will be satisfactory. For larger systems,
specifically 4 by 5 inch, these plates will in general not be
satisfactory.
3. Microflat glass which usually shows less than half a wavelength of
deformation is satisfactory.
4. If 4 by 5 inch commercial plates or film are used, then a liquid
. gate may be necessary.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
B-1-9
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Section B-2
The Effect of Aberrations Due to
Glass Plates in a Recognition System
Introduction
The apparatus used to make the two-beam spatial filter and the recognition
system both employ high quality lenses to obtain small well-defined diffraction
images of the point source.
The size of the pinhole used is nominally .001 inch or 25?, but the size of
the Airy disc of the image as measured by a microscope (10X objective and 10X
eyepiece) was measured as being 20?, showing that the two lenses are quite good.
However, the spatial filters we plan to use consist of emulsion deposited
on a transparent base of a certain thickness, and we may use liquid filled cells
up to an inch thick. Such a thick plate, placed in a convergent beam of light,
increases the size of the image, because rays passing through it at greater angles
to the axis come to a focus at a point on the axis farther from the lens than the
focus of rays passing through the cell at smaller angles. Thus the addition of a
thick plate in the converging case of rays causes rays from successively greater
zones in the converging cone to come to a focus at points distributed along the
axis, instead of all coming to a common focus. The envelope of such a cone of light
comes to a minimum diameter W and then increases. The object of this work was:
(a) to calculate the position, along the axis, of the focus
of rays of various inclination, and
(b) calculate or estimate the enlargement W of the focus spot
due to this, and
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5 n-2-2
(c) set up an experiment on the six-foot optical bench to check the
calculations.
Derivation of Equations
Figure 1 shows a ray of light, part of a converging cone, passing through a
flat plate of thickness t and index of refraction Ng and, because of displacement
in the plate, coming to a focus at e, instead of the original focus C. The displacement d
along the axis I is the distance between C and e. Distance j is the distance from the
plate to C, but as j doesn't appear in the equation derived below, its value doesn't
:111fect d.
From Figure 1, equation 1 can be derived: (see page 5)
d = t (1 cosi
Ng cos r
in which the value of r can be computed from Snell's law, viz:
sin r = (Na/Ng) sin i
H-
- I. a
(1)
(2)
Approved For Release 2005/05/02 : CIA-R924743B507
Figure L 0 3000416r-I
F
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Calculations
These tests were made on the small system, not the 20-foot system.
The small system has more nearly perfect lenses and operates at a higher
aperture ratio so the effects should be more easily seen. In this instrument
the convergent cone of light was 26 mm diameter at lens L-2, and converged
350 mm away. To select values for i, the lens was divided into three zones
of equal area and the three radii used to establish three values for i.
Equations 1 and 2 were used to calculate 3 corresponding values for d as shown
in Table 1, where other values assumed were:
t = 24 mm
Ng = 1.51
Na = 1.00
Table 1
Case Angle i d
1 2? 8' 8,112.1 microns
2 10 44 8,110.0 microns
3 10 14' 8,108.0 microns
B-2-3
Results of calculations tabulated in Table 1.
(a) The distance along the axis between the three foci of these
three sets of rays is only 2 microns.
(b) A further calculation of W, the enlargement of the focus spot,
showed W = 0.12 micron.
Such an enlargement is too small to measure, the chief purpose to be served
by an experiment was to look for an observable enlargement, which would check on
the calculations.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5 n_2_4
Experimental Setup
The six-foot optical bench was brought into the best possible adjustment,
with the laser beam passing through the pinhole purifier, through a collimating
lens L-1, through converging lens L-2, through an observing microscope having three
translational motions (from the Gaertner lens testing bench) and fitted with a 10X
objective and 10X eyepiece, and then through a neutral density filter and into the
observer's eye.
The thick plate, placed between L-2 and the microscope, was a high quality
glass cell filled with Decalin, a liquid with an index of refraction 1.51 matching
that of the glass in the cell. The cell was 24 mm thick.
Results of Experiment
The value of d was measured as 7.68 mm (instead of 8.11 as calculated in Table 1).
The probable cause of this discrepancy is that the value of Ng is not precisely known.
The image of the pinhole, in a well adjusted system, consists of central disc
(Airy disc) plus concentric diffraction rings. Some 837 of the energy lies in the
Airy disc, and its diameter is what was measured.
With cell omitted, disc diameter = 20[1.
With filled cell in place, disc diameter = 16 to 18[1.
This indicates that the aberrations added by the cell have a sign opposite to
residual aberration in the lenses; unless the apparent reduction in disc size is
due to inaccurate measurement.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
B-2-5
Experimental Setup
The six-foot optical bench was brought into the best possible adjustment,
with the laser beam passing through the pinhole purifier, through a collimating
lens L-1, through converging lens L-2, through an observing microscope having three
translational motions (from the Gaertner lens testing bench) and fitted with a 10X
objective and 10X eyepiece, and then through a neutral density filter and into the
observer's eye.
The thick plate, placed between L-2 and the microscope, was a high quality
glass cell filled with Decalin, a liquid with an index of refraction 1.51 matching
that of the glass in the cell. The cell was 24mm thick.
Results of Experiment
The value of d was measured as 7.68mm (instead of 8.11 as calculated in Table 1.)
The probable cause of this discrepancy is that the value of Ng is not precisely known.
The image of the pinhole, in a well adjusted system, consists of central disc
(Airy disc) plus concentric diffraction rings. Some 837. of the energy lies in the
Airy disc, and its diameter is what was measured.
With cell omitted, disc diameter = 204.
With filled cell in place, disc diameter = 16 to 18u.
This indicates that the aberrations added by the cell have a sign opposite to
residual aberration in the lenses; unless the apparent reduction in disc size is due
to inaccurate measurement.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5B-2-6
Appendix
Derivation of Equation 1
From Fig. 1 k = 1 sin (i-r)
Expanding k = 1 sin i cos r-1 cos i sin r
But 1 = t/cos r
t sin r cos i
k = t sin i -
cos r
By Snell's Law, if Na = 1 sin r = (sin Oft
k = t sin i -
t sin i cos i
Ng cos r
k = t sin i (1 - cos i
Ng cos r
By Fig. 1 k = d sin i
So d = t (1 - cos i
Ng cos r
(1)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
B-3-1
Approved For Release 2t1Y5t0SR92B:-elA-RDP781304770A002300020010-5
VIBRATION ANALYSIS OF 20-FOOT OPTICAL BENCH
The vibration characteristics of the 20-foot optical bench were investigated
by applying a variable frequency vibration force to the frame and observing the
resultant vibration motion. The frequency of the input force was slowly increase
and these frequencies recorded at which maximums in the vibration amplitude were
noted. In this manner the resonant or critical frequencies up to 10 kilocycles per
second were determined.
By holding the frequency constant at each of these resonant frequencies the
vibration amplitude distribution throughout the bench was determined. The resonant
frequencies and a sketch of this vibration amplitude distribution is shown in
Figure 1.
When the effect of the various vibration modes on the operation of the optical
system was studied it was found that only one at 45 cps was important. This is the
first, or lowest frequency mode, and consequently the one with the largest displacement
for the same energy input, of the modes that result in a distortion of the optical
axis from bending stresses in the bench.
It was observed that, at 45 cps, if the center of the bench vibrated over
.0004" peak to peak then the loss of light in the optical system was noticeable.
Since there is a resonant amplification of 20, the maximum acceptable floor vibration
at 45 cps without isolation is approximately 20 microinches peak to peak. However,
with the vibration isolators between the bench and the floor the maximum floor
vibration could be .001" (0.1g acceleration) peak to peak before the effects of the
vibration were noticeable.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
6.4 cps
cps
83 cps
12.1 cps
42.5 cps
45.0 cps
103.0 cps
118 TO
123 cps
180 cps
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5 11- 3-2
VIBRATION CHARACTERISTICS OF 20 FT. OPTICAL BENCH
VII:111A1 IUN AMI'Ll I UUt Mb I HUN I IUN
NOTE: fn fn = 8.3 cps WHEN TRANSFORMER MOVES IN PHASE WITH THE
FRAME AND fn =12.1 cps WHEN TRANSFORMER IS NOT MOVING.
RESONANT FREQUENCY OF THE IMAGING LENS ON ITS MOUNT,
0?w 10.
THIS MODE OF VIBRATION CONSISTED PRIMAIRLY IN HIGH AMPLITUDE
OF THE CROSS BRACES IN THE STRUCTURE OF THE BENCH.
FIGURE I
>4?
OTION
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
25X1 Note t
Report
During this testing program it was found that the only two parts that
vibrated independently of the bench were the imaging lens mount and the "noise
eliminator". Both of these parts were poorly supported. The imaging lens was on
the end of a thin rod and the noise eliminator was cantilevered from the end of the
laser on a rather insecure frame. Both of these parts were rebuilt in accord with
the general solid construction of the lens mounts and no further vibration was
noticed.
In general, the construction appeared insensitive to even very severe
vibration, the equipment continued to operate as a recognition system all during the
vibration tests and afterwards with no change of adjustment. So far as we can see,
vibration is not a problem with the type of construction used.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05p:4i g4-13Df781304770A002300020010-5 13-4-1
Beam Attenuation Methods
Introduction
The apparatus required to make a two-beam spatial filter (Figure 14 of
Reference 1) includes an attenuator to reduce the intensity of one of the light
beams. This investigation concerns a search for the best material to use for
this attenuator.
Desirable features of such an attenuator should include:
(a) the beam of light transmitted should be reduced in intensity
but otherwise unchanged. That is, aberrations or scattering
should be absent, or as small as possible.
(b) attenuation should be continuously variable, although for
some applications, small steps would be equally useful.
Other features, such as cost, weight, size, reliability, life expectancy
and maintainability depend too much on the specific application to be given much
consideration here.
Attenuators Investigated
(a) thin Wratten neutral density gelatin filters mounted in cardboard
holders without glass.ND=1.00.
25X1
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
B-4-2
(b) small (2 x 2 cm) Wratten neutral density filters, factory mounted
in glass. ND=0.3.
(c) Bausch & Lomb N.D. filters (2" x 2") made by evaporating metal
on glass. ND=0.30.
(d) continuous neutral wedge on 6" diameter circular glass plate,
5/16-inch thick. This was made by Eastman Kodak and cost $600.
(e) Polaroid type HN 38, plastic sheer: 12" x 12" x .032".
Sheet was unsupported and rather curled.
(f) Polaroid type HN 32, stiff plastic sheet. (Surplus pieces
selected for use with 3-D Color TV Display System.)
(g) Kodak Polaroid filter for cameras, called Polascreen. Here the
Polaroid is mounted between high quality glass plates.
Method of Investigation
The apparatus on our six-foot optical bench was set up as used for a 2-beam
recognition system (Figure 15 of Reference 1). The imaging lens, the object and
the spatial filter were removed, and an observing microscope having three trans-
lational motions (from the Gaertner lens testing bench) was set to look at the
source image located in the plane where the spatial filter was. Various
attenuators were placed in the beam, both at the image plane and nearer the
converging lens. Observations were made at maximum effective magnification, 100X.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
B-4-3
The method used to decide whether an attenuator would be satisfactory was
to observe changes in the size and shape of the 20 micron diameter airy disc of the
source image. These results are somewhat subjective, in that the degree of goodness
or badness of the image could not be given on a scale of numbers, but judgements had
to be made as to the general crispness and detail in the image and how these changed
as the attenuators were moved in and out of the light beam. Each time an attenuator
was moved in or out of the beam it was necessary
the picture
as sharp
making comparisons.
as possible, and
When a polarizer
this added
is
used as
to refocus the microscope to get
somewhat to the difficulty of
an attenuator in the apparatus
used to make a 2-beam spatial filter (Figure 14 of Reference 1), the system will be
degraded unless two polarizers are used, with the second one (the one farther from
the laser) permanently rotated around the optic axis so that its plane of polarization
is parallel to the laser's plane of polarization. The reason for this is because:
(a) when several of plane polarizers (such as Polaroid sheets) are
placed in series in a beam of light, the plane of polarization
of the light transmitted through all of them will be parallel to
the plane of polarization of the last polarizer.
(b) for the 2-beam spatial filter to be effective in doing its job
it must contain many fine interference fringes created by the
interaction of the upper and lower beams. The higher the contrast
in this fringe pattern the more effective the filter will be. In
order for two beams of light to interfere to form fringes their
planes of polarization must be parallel, and as they are made
increasingly non-parallel, the contrast between light and dark
fringes decreases.
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
B-4-4
(c) Therefore, for maximum fringe contrast, as i (b), the plane of
polarization of both beams must be kept parallel and the 2nd polarizer
must, because of (a) be fixed, and the 1st polarizer rotated to cause
variable attenuation of the beam.
In cases reported below a single polarizer was used (turned so its plane
polarization has approximately parallel with that of the laser beam) but in
actual attenuator two polarizers would have to be used with consequent doubling
it the scattering or aberration of a single one. Therefore, in case
two polaroids were used. (In cases e, f, & g this was not done as only a single
polarizer of each type was available
Results of Experiments
All of the attenuators operated satisfactorily when placed at the focal
point of the second lens, or within a few millimeters of the focal point. However,
if a dust speck should happen to fall on the surface of an attenuator so located,
Hle amount of light scattered from it would be large and hence such a location
might not be a good one.
As an attenuator was moved from the focal point to positions nearer the
tocusing lens where the beam cross section was larger, the performance fell off
in many cases. This is to be expected if an attenuator does not have sides that are
tlat, because lack of flatness will deviate a ray as it passes through, and the
deviation will vary from one point to another. If the rays pass through a small
;:rea of the attenuator the deviation will be more nearly the same for all the
,nys than if they pass through a large area.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
The results of the tests are tabulated below:
Material
(a) Gelatin
Small Area
of Beam
Large Area
of Beam
B-4-5
Remarks
Good Good Beam cross section can
be any size with only
slight image deterioratio
noted. Max. beam size
about 20mm.
(b) Small Wratten in glass Good Bad
(c) B & L 2" x 2" glass Good Bad
(d) EK Circular '..ledge Good Good Image deteriorates
noticably only when beam
cross section is 16 mm
diameter or 640,000 X
area when at focus.
(e) Polaroid Plastic Sheet
Type HN 38
Good
Bad
Sheet poorly supported
and rather curled.
Might be OK if flat.
(0 Polaroid Stiff Plastic Sheet
HN 32
Good
Bad
Beam diameter 2 mm when
image deteriorated.
(g) Kodak Polascreen - Polaroid
Good
Good
Quite good at all beam
between high quality glass diameters. Only had one
plates Polaroid so couldn't test
when crossed. Next item
indicates 2 would be OK
crossed.
(h) 2 Sheets HN 32 Good Less Good When beam diameter was
3 mm Airy disc was a
20 x 25 microns ellipse.
When 1st sheet was
rotated 90? image got
very dim but no change
in image quality was
noted.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Conclusion
B-4-6
Satisfactory attenuators would include:
1. A series of gelatin filters (item a).
2. A circular wedge (item d).
3. Two crossed polarizers - if mounted in flat glass plates (item g).
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Section B-5
A New Means for Attenuating or
Intensifying the Reference Beam
B-5-1
Introduction
In making spatial filters it is usually necessary to attenuate the reference
beam to obtain optimum interference in the desired diffraction order. We have
made a study of the conventional means such as neutral density filters and crossed
polarizers. There are also times when it would be desirable to attenuate the sample
beam, but this is not recommended because it is practically impossible to obtain an
attenuator that will not introduce undesirable phase variations. The only recommendat-
ion has been that the apparatus should be designed so that this never happens or we
should avoid large area objects which are likely to cause this situation. Needless to
say, these recommendations are not entirely practical.
Possible Alternative Attenuators
1. Iris on the Rayleigh Lens. An iris placed on the Rayleigh or reference
beam lens would allow control of the amount of light in the reference beam and act as
an attenuator, but it also changes the diffraction pattern of the point focus. This
point focus later becomes the recognition "spot", and hence changing this aperture
would alter the size and structure of the recognition image. Since it is normal to
attenuate the reference beam as much as 1:1000, the f value of the Rayleigh lens
would become about f 2000 and the central disk of the focus and later the recognition
spot would become large and diffuse. This solution appears impractical.
2. Vary Focal Length of the Rayleigh Lens. When the focal length of this
lens is changed the reference beam at the filter plane becomes larger or smaller
in diameter and hence its energy per unit area varies. This has some of the same
difficulty as (1) above. The aperture remains constant but the f value will decrease
with long focal lengths. This is not nearly so severe as in (1) and is a minor
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
B-5-2
objection. The major problem is finding a set of a dozen or more diffraction limited
lenses, each three-inch aperture with the required focal lenghs. Even if the lenses
were available, their use would be a considerable problem. Each time the focal length
of the Rayleigh lens was changed the position of the object would have to be changed
because the object and this focus must be in the same plane. In the "overlapping"
system which we use in the object is not in collimated light this would change the
object magnification and require a whole set of changes in the setup. This solution
also appears impractical.
3. Add a Weak Lens Beyond Focus. If a weak simple lens such as a spectacle
lens, is placed in the reference beam after the point focus it changes the diameter
of the beam at the filter plane. The light intensity varies as the square of the
diameter of the beam so that relatively weak lenses provide all the control required.
Since the beam is small in cross section close to the focus, only a small portion of
the lens is used and low-quality lenses are satisfactory. When negative lenses are
used they spread the beam and act as attenuators, positive lenses compress the beam
and increase the intensity of the reference beam which is better than attenuating
the sample beam in terms of exposure.
In a test, this means of attenuating the reference beam or increasing its
Lntensity appeared completely practical using a set of ordinary long focal length
single element lenses.
In the case where the sample beam had to be attenuated by a filter (a
practice that is not recommended) the exposure for a spatial filter was 4 minutes.
Using the same equipment, but increasing the intensity of the reference beam by the
auxiliary lens produced a better result with an exposure of 24 seconds.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
B-5-3
Conclusion
The use of an auxiliary positive or negative lens in the reference beam can
be used to attenuate its intensity or increase its intensity as required. This
is the first practical means we have found to increase the intensity of the reference
beam.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02se9A-gpVgB04770A002300020010-5B-6-1
Use of Glass or Film to Make Spatial Filters
Introduction
All the spatial filters mentioned in this or previous reports from this
laboratory were made on Eastman Kodak Spectroscopic Film type 649-F, 35 mm size.
So far as known we are the only laboratory using this material and most others
do not know it exists. The advantages of film are easy loading in a 35 mm camera,
easy processing, easy identification of individual exposures.
We have been criticized for not using glass plates as others do. The
disadvantages of plates are that it is impossible to buy them in the proper size,
so they must be cut in the dark. Unless done with great care the emulsion surface
can be damaged. Because cut plates vary in size and thickness they are difficult
to hold during exposure. We have found it much more difficult to design a device
to check the focus on these variable thickness plates than film.
Experimental Procedure
Spatial filters were made of the same object on both spectroscopic film,
plates 0.40 inch thick and microflat plates .25 inches thick.
The exposure time to obtain the same density in the image was different
for each material. This is normal in these special order emulsions. The emulsion
on film has always been slower by a factor of 2 to 10 times than the same type
emulsion on glass.
Examination of Images
Microscope examination of the images on the three substrates showed only
one important difference. The images on plates showed ghost halation images due
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
11-b-2
to reflection from the back of the glass. Glass plates are not backed with anti-
halation coating, though it is possible for the user to coat them himself. The
film is backed with a very efficient coating which comes off between development
and fixing.
Test of Filters
Three glass spatial filters and three corresponding film filters of the same
object were compared to a photograph of ten similar objects. No difference could
be detected between the performance of the two kinds of filters. The actual
performance of two of the film filters was better in terms of signal strength, low
noise and ability to reject similar targets, but this could be explained by slight
differences in density. The remaining pair showed no difference.
Conclusion
It is impossible to hold spatial filters made on film perfectly flat. This
appears to make little or no difference. Filters have been purposely bowed to a
fairly short radius and they still operated. So long as the bowing does not cause
appreciable foreshortening in one dimension or displace the filter from the plane
of best focus, the effect seems negligible.
Filters made on film do not show reflections from the rear side of the
substrate.
Glass filters appear to have no advantages for the present experiments.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Section B-7
Liquid Gate
B-7-1
Introduction
Most of the users of systems for optical processing of information in
coherent light use liquid gates to hold their object transparencies. The
reason usually given is that the liquid gate suppresses irregularities in
the glass or film support or in the thickness of the emulsion coating. Other
reasons we have seen in print are that liquid gates are useful to suppress the
wedge angle of glass plates which would otherwise deviate the diffracted light
beam from the spatial filter. Many thin glass plates also show appreciable lens
power, either positive or negative so that it is necessary to adjust the position
of the spatial filter axially to compensate for this change in focus. The liquid
gate should remove this effect also.
Finally, we have seen the statement that liquid gates were useful to suppress
the unwanted relief image of the transparency.
Not mentioned in any reports we have seen is the fact that a liquid gate
should suppress the diffraction effects of scratches and defects on the surfaces
of the emulsion and film support.
Liquid Gate Design
The gate used in these experiments was made with one fixed window and one
adjustable so that the wedge angle could be reduced to practically zero. The
windows were made from the best of several microflat glass plates and showed less
than one-quarter wavelength phase distortion over the central two-inch area.
Attempts to use volatile solvents as immersion liquids were not successful.
Evaporation at the surface resulted in cooling and convection currents so that
the liquid-filled gate showed very bad optical astigmatism. This was due to a
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
B-7-2
convection pattern showing descending cooled liquid at the center and two
rising columns of warmer liquid at the edges of the cell. The difference in
temperature was not measured,but the result was about five wavelengths of phase
distortion.
The only satisfactory liquid was mineral oil which is difficult to remove
from the negatives.
Experiment
An aerial negative showing airplanes and a fair number of surface scratches
was tested both in and out of a liquid gate filled with mineral oil. In several
measurements there was no appreciable difference in either the signal level or
the signal-to-noise ratio. Also, the surface scratches looked practically as
bright when immersed as when in air.
This was an unexpected result; in all previous experiments immersion had
suppressed the relief image and reduced the signal strength to about half its
value in air.
Conclusion
In spite of the many advantages claimed for liquid gates, we have been
able to obtain equally good results without them. In tests previously described
in an earlier report, we also made spatial filters from objects in liquid gates
and used spatial filters in liquid gates with no change in results.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5 B-8-1
Section B-8
Test of S.O. 243 Film for Spatial Filters
Introduction
This special order Eastman Kodak film was obtained in 35 mm size. This
is described by the maker as "aerial film with ultra-fine grain and high acutance".
The base is 5.25 mil triacetate without gelatin backing. This gray base does not
provide the same high transparency as the base on which spectroscopic film is
coated.
The sensitivity is panchromatic with extended red sensitivity. The spectral
curve is practically smooth between 5500 and 6200 angstroms shows a dip at 6500
and a final peak at 7000 angstroms. The limit of sensitivity is about 7150
angstroms.
The RMS granularity in D-19 developer is rated at .016, in D-76 developer
it is .013 for a density of 1.0
The resolving power for high contrast objects is as follows:
Developer Resolution
D-19 465 1/mm
D-76 525 "
For low contrast subjects 1.6:1 the resolution is:
Developer Resolution
D-19 205 1/mm
D-76 205 "
The ASA speed is rated at 1.6 which is 1000 times the rated speed of
spectroscopic emulsion of .0015. However, we find little relation between the
ASA rating and the effective exposure in laser light at 6328 A. For a particular
spatial filter the optimum exposure on 649-F was 50 seconds, the optimum exposure
on S.O. 243 was between .1 and .2 second depending on the development. This
amounts to about 300 times the exposure speed.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
B-8-2
It is possible to develop this film in D-19 for 5 minutes or D-76 for
8 minutes to obtain "normal" contrast. Both of these treatments appear to
produce equivalent results. It is also possible to develop in D-76 for 2 minutes
to obtain a very low contrast that requires doubling the exposure to obtain the
same density and to reach the optimum diffraction efficiency. This treatment
appears to improve the efficiency of the filters.
It would be desirable to have a fast film such as this available for spatial
filters of poorly transmitting objects. For example, the exposure time on
spectroscopic film for a 2 mm tank image is about 30 minutes. The only other
fast film we have used is Kodak High Contrast Copy film which has an ASA speed
of 32. For the spatial filter that required 50 seconds on 649-F and .15 seconds
on S.O. 243, the correct exposure on HCC film was 1 second.
Experimental Work
The modulation efficiency of a spatial filter is the important property in
determining the brightness of the recognition image. The optimum values we could
obtain on the three emulsions were as follows:
Relative Modulation Efficiency
Emulsion
Development
Efficiency
649-F
D-19,
5 min.
100
S.O.
243
D-19,
5 min.
37
S.O.
243
D-76,
8 min.
50
S.O.
243
D-76,
2 min.
75
H.C.C.
D-19,
5 min.
50
These results indicate that S.O. 243 is faster than HCC film and with proper
development can show a higher modulation efficiency.
Reconstructed Images
Some idea of the quality of a spatial filter can be gained from a microscope
examination of the reconstructed image. For the subject tested, there was little
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5 B-8-3
difference in the three films. The reconstructions from spectroscopic film were
slightly more detailed and crisper with practically no difference between S.O. 243
and HCC film.
Graininess
When spatial filters of identical density were examined under the microscope
the following results were obtained:
Apparent Graininess
Emulsion
Development
Graininess
649-F
D-19,
5 min.
Very low
S.O.
243
D-19,
5 min.
Low
S.O.
243
D-76,
8 min.
Same as above
S.O.
243
D-76,
2 min.
About same,
different pattern
H.C.C.
D-19,
5 min.
Larger than above
It appears that the resolution cut-off due to graininess of spectroscopic
film is 3 to 4 times as high as either SO 243 or HCC. The reported figures are
as follows:
Resolution
Emulsion Development Resolution
649-F
D-19,
5 min.
+ 1000 1/mm (?)
S.O.
243
D-19,
5 min.
465 1/mm
S.O.
243
D-76,
8 min.
525 1/mm
HCC
D-19,
5 min.
360 1/mm
Relief Effects
Emulsion relief effects contribute to the efficiency of a spatial filter,
but they are also believed to have undesirable phase effects. The noted relief
effects were as follows:
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP781304770A00230002R0045
Relief Effects
Emulsion
Develgpment
Relief Image
649-F
D-19, 5 min.
Very low
S.O.
243
D-19,
5 min.
Higher with some
roughness
S.O.
243
D-76,
8 min.
Still higher with
more roughness
S.O.
243
D-76,
2 min.
Less relief,
smoother
H.C.C.
D-19,
5 min.
Same relief as
(2) above, rougher
It is impossible to decide whether high or low relief is desirable, but
certainly roughness of the emulsion is undesirable since it contributes to scattering.
S.O. 243 is superior to HCC in this respect.
Conclusions
S.O. 243 film can be used to make spatial filters. It does not show the
extremely fine-grain and high modulation efficiency of spectroscopic emulsion and
therefore does not make a filter as sharp or efficient. However, it requires an
exposure only about one-three hundredth as long as would be useful for some objects.
Proper processing will reduce the surface roughness and apparent graininess to
acceptable values.
S.O. 243 appears to be superior to High Contrast Copy film in every respect
including speed.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010-511_8_5
Section B
Conclusions
I. Poor quality glass plates in sizes 4 x 5 inches or larger can upset the
operation of the system. At present this is no problem since we plan to
work chiefly with film.
2. The spherical aberration introduced by glass plates or liquid gates is
inconsequential in systems such as we have used.
3. Vibration effects can be avoided by proper design of the equipment.
4. Either of three methods can be used to attenuate the reference beam. The
one favored is the use of unmounted gelatin filters.
5. By adding an auxilliary lens it is possible to either attenuate or
intensify the reference beam.
6. The performance of spatial filters made on glass or film appears identical.
Film is much easier to use.
7. Liquid gates have shown no advantages in experiments where they were used.
We have obtained good recognition on aerial film with the film in air.
8. Emulsions other than slow spectroscopic 649-F can be used with comparable
results and much shorter exposure time.
9. The general conclusion is that all of the equipment and materials problems
considered appear to have simple solutions and are not limiting factors to
the operation of the process.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A0023000200c0-t4
Section C - Automation
It was established as early as 1963 that the spatial filtering process
was capable of recognizing objects regardless of their complexity in photographs.
Since that time a good deal has been learned about techniques and some data has
been accumulated on what the process can and cannot do. It appears to have
considerable potential, but it is of no economic value until it can be automated.
At the present time, using engineers to make the spatial filters, visually
align the equipment and detect the recognition spots by eye it is far more
expensive and time consuming to find an object in a photograph than a simple
visual search using a magnifier.
If the system is ever to be used economically it must be automated to
perform the search and recognition mechanically. There are many problems to
the automation of a device this complex and in this study only a few critical
items were considered.
Before these tests are described it is helpful to understand the present
development of the manually operated equipment.
This section contains the following:
C-1 Description of the Equipment
C-2 Accuracy of Filter Alignment
C-3 Orientation Tolerance
C-4 Scale Tolerance
C-5 Time Required to Make an Orientation and Scale Search
C-6 Use of Photoplastic Film to Make Spatial Filters
C-7 Feasibility of Real-Time Filter Generation
C-8 Television Readout of Recognition Signals
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A0023000200104
Section C-1
DESCRIPTION OF EQUIPMENT
Bench
The optical bench supporting the equipment used in these tests was designed and
25X1 built and consists of two assemblies. One is a welded steel
frame that acts as a rigid base, the other is a pair of accurate rails that form the
Locating surfaces for the bench carriages.
The frame was designed as a steel truss bridge 20-feet long, 4 1/2 feet high
and 18 inches wide. It was made in two sections so that it could be transported on
an elevator and bolted together on installation. It rests on four feet which are
placed on vibration isolation pads. The frame is so rigid that no weight we would
consider placing on it would deflect the frame as much as .001 inch. The mass of
the frame is about 5000 pounds which effectively absorbs all vibration with frequencies
above six cycles per second.
The rails are hardened and ground stainless steel rods 20-feet long supported
in such a way that they can be straight even though the frame is not. Figure 1
shows an end view of the rails. They are fastened to 13 hardened and ground steel
yokes which determine the spacing of the rails. One of these yokes is different
(not shown) and is bolted directly to the frame. The other 12 yokes are supported
on adjustable screws and also laterally adjusted by two other screws between angle
plates. To facilitate initial adjustment, each yoke is held to the frame by a long
bolt in the center which passes through a clearance hole in the frame and terminates
in a coil spring and nut.
The rails were adjusted by autocollimating to a mirror placed on a carriage
sliding on the rails. Vertical and lateral adjustment of the 12 sprung yokes was
made until the rails were straight to at least .001 inch.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-1-3
FIGURE 1 Clamping Arrangement
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-1-4
Between the rails and attached to the yokes in a commercial steel shape
known as a "Unistrut". The carriages are clamped to the rails by a knurled-
head screw which passes through the carriage and engages a special nut in the
Unistrut. Large carriages have two clamps, small carriages have one.
Laser
The laser used in these experiments was a
Model 5300 D.C. 25X1
excited neon gas laser. The nominal output was 10 milliwatts, but more than 5
was seldom obtained. Since operation in the principle mode is necessary for making
spatial filters and no laser is completely stable, it would be desirable to use a
laser with externally adjustable mirrors.
Laser Mounting
The laser is supported at its two ends. Both of these supports should be
made adjustable to align the laser beam with the axis of the optical bench. Figure 2
shows the front end of the laser supported on two adjustment screws and retained by
a sprung pin at the top. The laser we used was stable for the first six months of
use, but thereafter required frequent adjustment.
Voltage Regulator
Not shown, is a Sorenson voltage regulator used to stabilize the laser. This
unit was necessary to obtain a constant laser power output, which before regulation
varied plus or minus 20 percent in a short term interval.
Noise Eliminator
This name is given to the device that brings the collimated laser beam to
a focus and removes the "noise" caused by dust on the laser windows, mirrors and
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
FIGURE 2 Laser Mounting
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5 C-1-6
the focusing lens. Figure 3 shows this assembly which consists of a negative
lens which spreads the 1.9 mm diameter laser beam so that it fills the back
aperture of the focusing lens. This lens also improves the energy profile of the
normally gaussian distribution of energy in the laser beam. The focusing lens is
a Unitron 10X microscope objective in a focusing, centering mount. The laser
beam is brought to a "point" focus on a .001-inch aperture in a very thin nickel
foil. This is also mounted in a centering mount which is not necessary.
The combination of adjustments allows the laser beam to be focused exactly
on the aperture and centered in it.
At
When experimental work is done where it is necessary to focus the intense
laser beam by eye and also take diffraction photographs on extremely slow film
it is necessary to have some control over the intensity of the light. This is
done with a rotating disc containing different neutral filters. The disc in use
has eight holes, one of which is open and used for all photographic work. The
other holes contain Wratten gelatin filters in various densities from .3 to 5.0.
This disc can be remotely stepped in either direction by a double-acting rotary
solenoid. The two control switches are in a small box which is normally at the
opposite end of the bench and which also contains eight indicator lights that show
which filter is in the beam. This information comes from an eight pole rotary switch
on the same shaft as the disc.
Figure 4 shows the disc at left, the two in-line rotary solenoids and the
switch all on the same shaft. The box at right contains the two push button switches
and indicator lights.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
FIGURE 3 Noise Eliminator
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
FIGURE 4 Light Attenuator
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
1-9
Light Monitor
Figure 5 shows a device used to keep the laser beam centered on the
collimating lens. The lens is in the mount at the right side of the picture.
Around the lens is a ring supporting four small cadmium sulphide photo cells.
These are connected to four meters to read the intensity at the Top, Bottom,
Right and Left sides of the lens. Initially, the laser is adjusted so that all
four meters read the same, plus or minus 2 units. If the laser shifts or the bench
warps the light spot will move off the lens and the light balance will be changed.
These meters are normally at the opposite end of the bench and are consulted every
time the measurement of a recognition spot intensity is made. Before this unit
available the beam would shift without our knowledge and assuming it was shifted
upward; then recognition spots obtained in the lower half of the picture would read
abnormally low. If all the meters read practically the same it can be assumed rnat
the object transparency is uniformly illuminated.
Lens Mounts
All the lens mounts are aluminum castings similar to the one shown in Figur
Not shown in this view are the adjustments. The lens is in a cell which is mounted o
a plate which fastens to the aluminum casting. The holes in the plate are oversize
so that the lens can be centerd on the optical axis. The lens cell also incorporate,
three push-pull screws which can be used to tilt the lens in any direction to square
it with the optical axis.
Rayleigh Lens
This lens is used only when spatial filters are being made and it then occupies
the top half of the beam as shown in Figure 6. When not in use the lens is swung
upward on a pivot out of the optical path. The lens has a mask around it to stop
stray light which might pass around it.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
0-C
ICROAMPERES 50
20 30
D-C
ROAMPERES 50
?,.?.
FIGURE 5 Light Monitor
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
FIGURE 6 Rayleigh Lens
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-1-12
Object Holder
The holder for 4 x 5 inch glass plates is shown in Figure 7. The plates
are retained between plastic supports and spring clamps. Other kinds of adapters
can be used for film and the liquid gate.
The holder can rotate indefinitely in either direction on a light duty ball
bearing. It is driven by a pulley around the outside and an "0" ring used as a
rubber belt. The small 10 watt reversible clock motor at left also has a pulley
for the rubber belt. The clock motor is one of a series and can be interchanged
to obtain different speeds of rotation. The small black box on the right contain!'
the controls and is normally at the end of the bench. The lower switch has three
positions, clockwise, off and counter-clockwise.
The entire carriage can be translated on an auxilliary set of rails 3 feet
long. The rails are precision ground stainless steel rods running in three ball
bushings. The position of the carriage is determined by a ball-bearing lead screw
which is driven by a pulley and belt arrangement from a reversible clock motor.
This motor is also replaceable with other motors having different gear ratios. Two
adjustable limit switches are provided to reverse the motion of the carriage. These
switches are provided to reverse the motion of the carriage. These switches are
attached to an auxilliary Unistrut shape between the rails. The switches can be
adjusted so that the excursion of the carriage is the full length of the rails or
only a short distance. The upper switch on the control box has two positions, on
and off. There is also a manual reverse button on the left side of the box to reverse
the direction of travel.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
,l11111i111111
St 40 10
FIGURE 7 Object Holder
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-1-14
These two motions, rotation and traverse are used to provide the orientation
and scale search of the object photograph.
The rotation of the object can be read to the nearest degree from the
scale at the top. The translation and hence the relative magnification is read from
an index mark on the moving carriage and a stationary steel tape which does not
show in the photograph.
Reference Beam Attenuator
When spatial filters are being made the object holder shown in Figure 7 is
used to support the transparency of the object in its lower half. The upper half
of the aperture may contain a metal plate with a small hole which acts as a "noise
eliminator" on the reference beam. In many cases it is necessary to reduce the
intensity of the reference beam to obtain interference fringes in the desired
diffraction orders. This is done by inserting a Wratten neutral density gelatin
filter in the beam. It is essential that this filter be as close to the focus as
possible. A means for doing this is shown in Figure 8. A metal plate painted
light gray in the photograph is inserted against the back of the object transparency.
This locks in place with two rotating sprung cleats, the dark colored objects in the
photograph. Gelatin filters mounted in cardboard frames are inserted under the
spring clips. In the figure a filter labelled N.D.0.50 is in place in the upper ,r
reference beam and a small object transparency is in the lower sample beam.
Spatial Filter Holder
This device has precise X and Y motions controlled by large micrometer heads.
This part was purchased as a stage for a small toolmaker's microscope and was
intended to operate in a horizontal position. The two tension springs at the top
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
FIGURE 8 Reference Beam Attenuator
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C- 1-16
in Figure 9 are to take part of the weight off the lower micrometer.
The center part of this device can be rotated and the angle read in the
curved window at the left by means of a vernier and magnifier. The angle can be
measured to at least 1 minute of arc.
The spatial filter is attached to a metal plate with masking tape and the
metal plate fits the aperture in the stage. Two spring clips retain the filter
in use. The spatial filter shown in the figure is typical for a fairly simple ob't
Imaging Lens
This lens is installed just behind the spatial filter and forms the images
of the recognition spots. When the filter is changed or adjusted the lens 3wings
up out of the way on a pivoted mount. Figure 10 shows the lens in its normal
position.
Viewing and Photometric Microscope
The recognition images are viewed with a low-power microscope shown in Figure
which can also be used to measure their intensity. The microscope has a special
mount which allows it to tilt so that it is in line with the off-set first order
recognition image. The microscope also has rack and pinion motions in X, Y and Z.
At the center of the field of view there is a single glass fibre built into the
eyepiece. This fibre is bent at a right angle and makes contact with a 3-foot long
multi-fibre light conductor. This goes to a phototube, not shown. The output of the
phototube
the meter
and meter are commercial items purchased from the
is read on the meter shown which has several scales. The power supply for
is battery operated to avoid live current variations. The entire microscope
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
FIGURE 9 Filter Holder
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
FIGURE 10 Imaging Lens
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
FIGURE 11 Photometric Microscope
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C- 1-20
There are several parts not shown in the figures which do not need
explanation.
The camera used to make spatial filters is a 35 mm reflex camera without
a lens. The reflex mirror is used to find the point of best focus and the normal
magnifier is replaced by one of higher power.
The same camera is used to photograph recognition images.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5C-2-1
Section 0-2
Alignment Tolerance of Spatial Filter
Purpose
It is necessary to remove photographic spatial filters from the equipment
for processing and then to replace them so that the image of the zero order is
aligned with the optical axis. So long as this is a hand operation, the alignment
tolerance is of no importance because the filter can be aligned by eye using a
microscope and a suitable positioning device. If the recognition process is ever
automated, then it will be necessary to have some means to do this automaticaily
and to know the tolerances to which this device must operate.
Possible Automation Devices
In its present state of development the recognition process requires three
separate optical systems, or with some loss of convenience these three can be
incorporated in one by interchanging parts. These three systems are:
1. Filter making system using the Rayleigh lens to produce the
two-beam filter.
2. Filter testing system to view the quality of the reconstructed
image and measure its relative modulation efficiency.
3. Recognition system with equipment for measuring the intensity
of recognition spots.
Thus, in the present equipment, the filter is made in one instrument and used in
another so that some means would have to be devised for removing the filter,
processing it, and returning it to an equivalent location in a second instrument
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-3
If we are willing to ignore this advice and can build a single unit system,
then the problem of alignment tolerance disappears. The filters would be exposed,
and if silver photographs would be rapid-processed in place and used there until
discarded. Photoplastic filters could be processed in place and erased when no
longer needed.
Considerations of automation seem to lead to two alternatives:
1. Separate instruments for making and using filters. Some mechanical
device such as an accurate frame to allow the filter to be transferree
from one instrument to another without realignment. Such a system
appears more flexible and allows the user to establish a "filter bank".
2. A single instrument which processes and uses filters in one place.
At the present time our thinking favors the first alternative and hence the
necessity for determining the alignment tolerance.
Experimental Approach
An obvious and simple means for determining alignment tolerance is to measure
the intensity of the recognition signal and the signal-to-noise ratio when the filtr
is properly aligned and with different known amounts of misalignment. The problem
is to choose a typical object and a 'typical filter.
Experimental Equipment
Spatial filters of various vehicles and other objects
were carefully aligned with the diffraction images of photographs containing these
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
A rsvss......ehrl ?-???? DnInnca 9nrigingin9 ? rha_pnp7gR0a770A002300o2on1 0-5
J
)
.
I I 1
EFFECT OF DISPLACEMENT
. IN X OR Y FOR TANK
)
I
.
,
1
,
.
.
.?
,
.
.
i
.
NOISE
LEVEL
?
_
c cAl.2
,.. ,
...
_
8(
7
3
20
10
0
2 3 4
Approved For ReligrellgAffir: 8K-N648[9399)0102300020010-5
FIGURE I
5
6
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-6
The part of the curve from the plateau down to the noise level is practically
straight. The interpretation is that for alignment tolerance beyond the plateau,
the loss of signal strength will be linear. At its lower end the curve joins the
picture noise level in a gradual transition.
In every case we tested, the noise level did not change appreciably during
the misalignment. It is quite possible to have a major misalignment of the filter
which is much greater than any shown in the Figure 1 that can result in an increased
noise background. Measurement of the noise in this case can be a difficult problem.
Sometimes the recognition spot breaks up into several bright spots, or in some
cases individual features of the object are locally recognized giving rise to a
collection of scattered minor recognition spots. It could be argued that these
are also "noise", but we have felt that if they were within the area occupied by
the image of the object they were not background noise or false alarms and could be
neglected. Actually, these confusing conditions did not arise in any of the tests
designed to measure the alignment tolerance and therefore do not affect the inter-
pretation of the data. In all of these tests, when the filter was misaligned from
its proper location all that happened was that the recognition signal remained a
single, practically round spot, but decreased in intensity.
Our conclusion, for the tests performed, is that alignment error results in
a decreasing signal and a constant noise level and therefore a predictable decrease
in signal-to-noise ratio.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-7
It would be possible to continue these tests for objects of different sizes
and shapes and finally to generalize some conclusions for the sizes of objects we
might expect to search for in aerial photographs. These results are reported at
the end of this section.
An experimental test was made to determine if the direction of the interference
fringes had any effect on the relative x and y alignment tolerances. Normally, our
system produces vertical shear between the reference and sample beams so that the
fringes in the filter are horizontal. Also, it is our habit when making a filter
to arrange the object square and with its long axis horizontal. It was realized
that in some cases it would be advantageous to depart from this simple procedure.
Particularly, in the case of a multiple filter for several similar vehicles it
would not be desirable to arrange the vehicles parallel or in a neat square
formation because this would result in a strong diffraction pattern for the
arrangement. Since the purpose of the multiple filter is to detect individuals
and not groups arranged in a particular pattern, such a filter would be less
satisfactory than one in which the pattern was suppressed. Our means for sub-
ordinating this pattern is a random arrangement of vehicles. In this case the
different vehicles produce diffraction patterns at random angles and with
interference fringes at random angles.
Therefore, since this case appeared to be a real one, a test was made of
the displacement tolerance of a set of filters of the same vehicle in which the
interference fringes were parallel to the long dimension, perpendicular to the
long dimension, at 45 degrees and at a random angle which turned out to be 32 1/2
degrees. As might be expected, there was no measurable difference, so that as
a conclusion we can state that so far as alignment tolerance is concerned, it
makes no difference how the object is oriented when the filter is made.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-8
Analytical Approach
Another means to determine the alignment tolerance of a spatial filter is
to analyze a typical example to find the function of the filter elements and to
observe the effect of displacing these elements from the diffraction image. This
analysis should result in a set of simple formulas which would allow the operator
to calculate the tolerance for any possible situation.
ANALYSIS OF A SPATIAL FILTER
The problem is; can the various "spots" in a spatial filter be related to
features in the object from which it was made? For very simple objects such as a
slit or a round hole, the diffraction pattern is easy to predict and the analysis
of the spatial filter presents no problem. For a complex object, the hologram is
too complicated to analyze by eye. The object selected for study was a model M-48
tank. As seen in a vertical photograph the outline of the tank is a rectangle. The
long sides of the rectangle are straight, but on closer examination the ends of the
tank are not, one is rounded, the other is an offset straight line. The next major
features are a turret which is approximately elliptical in outline and a gun barrel
which is parallel-sided. The top surface of the tank also has cleats and other details.
Spatial filter Number 1077, which would recognize this tank or reconstruct
a good image was examined. It consisted of 23 vertical orders of diffraction and 20
horizontal orders of diffraction or about 166 spots since each of the horizontal
orders had two or more satellites. There were also sections of elliptical curves
and curved lines intersecting the vertical orders.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Tank Part 1
Tank Part 3
C-2-9
Tank Part 2
Tank Part 4
Tank
Figure 2 - Tank Parts Used in Spatial Filter Synthesis
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-10
In this experiment the filter was synthesized from its elements. Various
simple parts of the tank, the rectangular outline, the turret and so forth were
separately made into spatial filters with the idea that this would aid in identifying
the elements responsible for these various features. This technique is not foolproof,
as it may turn out that the diffraction pattern of two objects together is quite
different from the two separately.
SYNTHESIS OF A SPATIAL FILTER
Four India ink drawings of tank parts were made to the same scale. These are
shown in Figure 2 and show the following features:
Part 1 - Rectangular outline only, no details.
Part 2 - Outline with straight long sides and shaped ends.
This is the entire outline of the tank.
Part 3 - Rectangular outline with 12 cleats on the surface.
These cleats were considered to be the chief periodic
elements and the most conspicuous features of the tank
body.
Part 4 Outline of elliptical turret and gun only. The turret
is probably the most conspicuous feature of the tank
except its outline.
These four drawings were reduced to transparencies the same size as the
transparency of the tank used to make spatial filter Number 1077. Each of these
transparencies was then made into a spatial filter. The following results were
obtained:
Part 1, (Rectangular outline of tank only). The filter showed 23 vertical
orders of regularly spaced spots, each spot decreasing uniformly in density. Each
spot (order) was practically round and the spots were in a straight line. This
filter is shown in Figure 3a.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-11
Fig. 3a - Tank Part I
Fig. 3c - Tank Part 3
Fig. 3b - Tank Part 2
Fig. 3d - Tank Part 4
Fig. 3e - Tank Part 5
Figure 3 - Spatial Filters of Tank Parts
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-12
In the horizontal direction there were 25 orders of regularly spaced spots,
each spot decreasing uniformly in density. Each spot was oval in outline. The
entire pattern was rectilinear and symmetrical.
Referring to the description of spatial filter Number 1077 for a complete
tank, it could be seen that the rectangular outline is the chief feature, and that
the filter made from this outline only contained some 96 spots of which 86 compared
to 86 spots in the filter for a complete tank. This left some 80 spots to account
for.
Part 2, (Outline of tank with shaped ends). This filter, like Part I showed
23 vertical orders of regularly spaced spots, each spot decreasing uniformly in
density. Each spot was practically round and the spots in a straight line. These
vertical orders can be attributed to the diffraction of the long parallel sides of
the tank and therefore are the same for Parts 1 and 2. This filter is shown in
Figure 3b.
In the horizontal direction there were 13 orders or regularly decreasing density.
Each order was composed of three to five spots, the central three of which were most
dense. None of these spots was round, but varied from oval to banana-shaped. The
spots were not in alignment, but were shifted in their relative positions. The
pattern was essentially rectilinear and symmetrical.
Comparing this filter to Number 1077 for the whole tank it was found that
about 130 of the spots could be accounted for. This indicates that the outline
is the most important feature of the spatial filter and there were approximately
36 spots left to account for all other features.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C,2-13
1211_1, (Rectangular outline with cleats on top surface) like Parts 1 and 2
showed the same 23 vertical orders of regularly spaced round spots. This filter
is shown in Figure 3c.
In the horizontal direction there were 23 irregularly spaced spots (the cleats
are irregularly spaced) which tended to be oval. Different spots had two, four or
six satellite spots. Between some sets of spots were dog-bone shaped spots. In
general the order was as follows:
Order
1, 2
Number of Satellites
two
3, 4 four
5 two
6 to 10 six
11, 12, 13 none
14, 15, 16 four
17, 18, 19 two
20 none
20 to 23 two
This pattern is no doubt due to coincidences between diffraction patterns
formed by individual pairs of cleats.
A new feature was that the density of the spots now did not decrease regularly
as in the former simple cases. There was a maximum at about the 2nd spot and the 8th.
Again this is probably due to coincidences in the spatial frequencies in the object.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-14
The total number of spots in the spatial filter for Part 3 was about 184,
or more than can clearly be seen in the filter for the entire tank. This indicates
that the drawing showed the cleats in higher contrast than they appear in a photograph
of a tank, and therefore their diffraction pattern has more energy and is capable of
producing a larger number of orders. In general, the distribution of light in the
spots and their satellites is much the same as the spatial filter for the tank and
the differences between the filter for Part 3 and the whole tank are rather minor.
It is possible to pick out about 12 spots in the tank filter which cannot be accounted
for in Part 3. Therefore, we may say on a very simple basis, that there are 12 spots
left to account for due to the turret and gun.
Part 4, (Turret and gun only). This spatial filter required a longer exposure
than the previous objects. Parts 1 to 3 were all of the general tank outline and had
an area on the transparency of about 15 sq. mm. The transparency of the turret and gun
only had an area of about 5 sq. mm. The required exposure was about 50 percent longer.
This part makes a very peculiar hologram. The turret as seen from above is elliptical
in outline and produces a diffraction pattern that consists of concentric ellipses of
diminishing density. If these ellipses are examined carefully they are seen to actually
consist of broken arcs of circles which the eye joins together to make an ellipse.
Apparently, the process analyzes the elliptical curves into a number of circular
elements and each of these circular elements produces a bull's-eye pattern. The only
unusual feature is that there were apparently only 5 orders of diffraction produced
by the turret while the tank outline produced 23. The reason is probably the difference
in light distribution. In the case of a rectangle the diffraction is in the form of
spots in which the diffracted light is concentrated in a small area about .2 mm in
diameter or an area of .03 sq. mm. In the case of the elliptical turret the diffracted
light is in the form of "rings", the average diameter of each increasing with the order.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-15
The area of the 3rd order "ring" for example is about 1.77 sq. mm which is 60
times the area of the spot produced by the rectangle. If the intensity is the
same, the exposure will be distributed over a larger area and will be much less.
This filter is shown in Figure 3d.
The gun barrel in Part 4 is parallel-sided and therefore acts as a slit in
the negative and should produce parallel diffraction lines as long as the gun.
The conclusion thus far is that the turret alone should produce a set of concentric
ellipses, the gun alone should produce a series of parallel lines. However, the
gun plus turret produced a very different pattern which consisted of four ellipses
intersected by steep, nearly vertical lines at an angle of about 14 degrees. There
were about 5 of these orders due to the gun in the spatial filter. This is a case
where the sum of two diffraction patterns produced a pattern different from either
component.
Spatial Filter of Complete Tank
We can now look at the filter for the tank as shown in Figure 3c and recognize
all of the elements discussed.
Vertically there were 23 orders of diffraction, which did not decrease
regularly in density, but followed the following pattern:
Order Density
1 to 5 Strong
6, 7 and 9 Fairly Strong
8 Weak
10 to 16 Weak
17 to 23 Very Weak
Undoubtedly, the variations in density were due to the intersection of the ellipses
and slanted lines of the turret and gun. If these were in phase they reinforced the
vertical sWtpeovsiefd ggP Ifei
p egisies2enagrO2784VDMIEWa0g366162(384-%.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-16
The vertical spots were not all round. Orders 6 to 23 were practically round,
but 1 to 5 were distorted. This is also due to intersection of the ellipses and
slanted lines with the spots.
The vertical spots were not in a straight line. Orders 1 to 5 and 9 were
approximately straight. Orders 7, 10, 12, 14 and 16 were deviated to the right.
Orders 6, 8, 11, 13 and 15 were deviated to the left. This was probably due to phase
interference between the turret-gun combination and the tank outline.
There were 20 horizontal orders which did not decrease uniformly in density.
The progression was as follows:
Order Density
1 to 5 Strong
6 to 11 Decreasing Uniformly
12 to 16 Stronger
17 to 20 Decreasing Rapidly
This can be explained as due to the irregular spacing of the cleats interfering with
the odd-shaped ends of the tank.
Each of these horizontal orders was composed of from 3 to 5 oval or banana-
shaped spots believed to be due to the cleats and one curved end of the tank.
There were four orders of broken ellipses due to the turret and five orders
of slanted vertical lines due to the gun which intersected the 4th, 6th, 8th and 12th
vertical order of spots.
Taken together, it was possible on a simple basis to explain the presence of
every spot in the spatial filter of the tank from the four spatial filters of the
tank parts.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-17
ANALYSIS OF A SPATIAL FILTER
In the previous section a spatial filter was studied by synthesis; that is
to say, putting it together from its elements. In this way it was possible, to
some extent, to account for the presence of the 166 prominent spots and some of
the structure. There is a lot that this study did not tell and some of these
factors will be discussed in this portion which describes an analysis based on taking
the filter apart.
The actual means by .which this was accomplished was to successively mask
different portions of a spatial filter of a tank using small pieces of opaque
masking tape which were applied directly to the filter under a microscope. At each
step the partially masked filter was reconstructed and this image photographed.
The intensity of the reconstructed image was also measured. Finally, the masked
filter was used to recognize a tank and the intensity of the recognition spot was
measured. In every case the intensity of the reconstructed image and recognition
spot were comparable within the accuracy of measurement so that only one set of
figures is reported.
No tests were made of the noise background because, so far as could be seen,
it remained unchanged in all the tests. No tests were made of the properties of
the masked filters to distinguish between two similar objects because this was
outside the present scope which was simply to determine the origin and function
of each of the obvious elements, of the filter.
General
The filter contained 23 vertical "spots" representing diffraction orders of
the "slit" produced by the two long straight sides of the tank. It is fairly clear
from a microscopic examination of the filter that many of these spots are doing
little or no work in the recognition process. The first four orders are so much
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-18
overexposed that the bright lines in the interference fringes have irradiated or
spread to join the next bright line so that the dark fringes have disappeared and
the order is simply a solid black spot. Such a spot cannot produce diffraction or
contribute to the reconstructed image of the filter or the intensity of the
recognition image. These orders are said to be "plugged".
It is also apparent that the orders above the 10th are underexposed. The
light amplitude across a set of interference fringes is ideally sinusoidal and the
best diffraction appears to result when the exposure is such that the bright and
dark fringes are recorded as equal in width. As the exposure decreases, only the
most intense center portion of the bright fringe is recorded and therefore the
grating consists of narrow opaque bars and wide transparent bars. These gratings
appear to be much less efficient than the ideal grating with approximately equal
width bars.
Finally, as the exposure decreases still further the bright fringe does not
contain enough energy to completely blacken the film and the result is a grating
made of thin gray bars. The efficiency of these gratings appears to be negligible.
It was therefore anticipated that a fair number of the spots could be masked
Or removed from the filter without producing any visible effect.
This simple concept is complicated by the fact that the light in the various
vertical orders does not decrease regularly or according to any known function.
Measured values of the light intensity in one set of vertical "spots" was as follows:
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 :
Order
CIA-RDP78604770A002300020010-5
C-2-19
Intensity
0
100
1
3.0
2
1.06
3
.46
4
.40
5
.29
6
.11
7
.12
8
.078
9
.10
10
.058
11
.044
12
.037
13
.033
14
.033
15
.018
16
.024
17
.010
18-23
.010-.004
It can be seen that these orders do not decrease in decades or even uniformly.
Masking the Vertical Low Orders
As just described, the first four vertical orders appear to be completely
plugged and therefore non-functional.
Orders Relative Intensity Appearance
Masked Reconstructed Image Reconstructed Image
1-4 .98 No difference
. 1-5 .97 No difference
1-8 .66 No difference
These results indicate as predicted, that the first four orders are non-
functional, the 5th is only slightly useful and the 6th, 7th and 8th together
contribute about one-third of the energy transmitted by the filter for this
particular kind of information.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-20
Masking the Vertical High Orders
As described earlier, the higher orders are so weakly exposed that the resulting
gratings are inefficient. It should be possible to mask a good number of them with
negligible loss.
Orders Relative Intensity Appearance
Masked Reconstructed Image Reconstructed Image
23-15 1.00 No difference
23-13 1.00 Outline appear slightly
rough
23-10 .98 Outline is rough, some
loss of gun image
23-5 .10 Nearly total loss of
information on tank
sides and gun
The above results show that vertical orders 13 to 23, while distinctly present
in the filter are practically non-functional and orders 10 to 12 diffract little
energy. The fact that the outlines become rough as the higher orders are suppressed
is predicted by Abbe's theory of the reconstruction of a diffraction image which
states that resolution suffers if any of the higher orders are lost. Loss of
resolution in this case results in a wavy or ragged outline. The reconstructed images
are shown in Figures 4b and 5a, b, and c.
Conclusion on Vertical Orders
Taking these two experiments together it can be seen that so far as energy is
concerned, that only the 6th to 9th orders are efficient at diffracting light and
therefore, taking the average, this is a 7th or 8th order diffraction filter. It
would appear that a much simpler filter showing only the four orders from 6th to 9th
would probably be nearly as effective and certainly a filter showing only the eight
orders from 6th to 13 would be as effective as the complete filter. This leads to
the conclusion that it might be possible to make artificial spatial filters by
photographing drawings of appropriate "spots" composed of black and white lines.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
4a - No Vertical Orders Masked
4c - 1st to 4th Vertical Masked
C-2-21
4b - 23rd to 15th Vertical Masked
4d - 1st to 5th Vertical Masked
4e - 1st to 8th Vertical Masked
Figure 4 - Effect of Masking Horizontal and Selected Vertical Orders on a
Spatial Filter
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
C-2-23
It also shows that it should be possible by some process of composite photography
to produce a spatial filter in which all or most of the 23 orders were effective
at producing diffraction. If this were achieved, then the filter would be about
5 times as efficient as the present filter having only about four function orders.
There is also the possibility that a spatial filter made from a high-contrast drawing
would be more efficient than one made from a photograph.
Masking Horizontal Orders
The spatial filter shows some twenty horizontal orders of diffraction, most
of these orders having two to six satellites.
The results of masking these were as follows:
Orders Masked
Relative Intensity Appearance
Reconstructed Image Reconstructed Image
All Total disappearance of
details crosswise the tan!
20-8 .97 No difference
20-5 .94 Little difference
20-4 .90 Details rough
1-2 .80 Little difference
The conclusion from these tests is that only the 2nd to 5th orders are effective
in producing diffraction and the remaining orders are not serving any useful purpose.
Reconstructed images are shown in Figures 4a and 5d.
General Conclusions
While the filter examined contains some 23 orders of diffraction and some 166
diffraction "spots", only about four or five of these orders are doing most of the
work and only about 32 of the spots are effective in recognizing the rectilinear
details. This study did not cover the 12 spots which represent the turret and it is
not known how many of these are effective.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020016i- 2 5
441111-- 0008
RELATIVE OVERLAP OF SPATIAL FILTER AND HOLOGRAM OF TANK
FIGURE 6
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
100
90
80
70
30
20
10
Approved For Release200
: 4
? , ? ' 7
, 70A00
3000200110 5
?
?
EFFECT
DISPLACEMENT
I I
OF VERTICAL
FOR
A SLIT
.
.
FILTER
1217
0
0
o
?
,
0
.
.
.
.
_
R
15 12
9 6 3 0 3 6 9 12 15 11
DISPLACEMENT IN MILS (0.001")
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
FIGURE 7
I0
9
8
70
30
20
10
0
0.5 04 03 02 0.1 0 0.1 0.2
DISPLACEMENT IN MILS (0.001")
Approved For Release 2005/05/Of! Ilk%)1g78B04770A002300020010-5
Approved
For Release
2005/
s-
02 : CIA-RDP78:C4770A0MOR2PCOFF-4i0RIZONTAL
?
DISPLACEMENT
I I
FOR
1
A SLIT
.
FILTER
1217
.
.
.
.
.
.
03
04
05
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-2-29
Conclusions on Spatial Filter Alignment Tolerance
1. The curve of recognition signal strength as a function of x or y error
in alignment of the spatial filter is practically the same shape for any
object that we would consider a "target".
2. This curve has a practically flat plateau at zero error and linear slopes.
this means that small errors are insignificant and for larger errors the
loss of signal is proportional to the displacement error.
3. For objects in the size range which we would consider as detectible "targets"
an error of .001 inch results in negligible signal loss.
4. It should be possible to build an automated device to handle and change
spatial filters to an accuracy of .001 inch.
5. A simple geometrical explanation for the observed error curve has been
derived.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-3-1
Section 0-3
Rotation Tolerance of Spatial Filter
Purpose
If spatial filters are removed and replaced in the system or orientation
alignment is a problem, or if the objects have to be scanned in orientation, then
it is necessary to know the orientation or rotation tolerance of the filter.
Effect on Automation
If an automatic orientation search must be conducted it is necessary to know
this tolerance to set a timing cycle.
Experimental Approach
A number of objects such as large and small vehicles, and characters were
aligned with their spatial filters and the intensity of the recognition signal
measured. The object or the filter was then rotated a small amount and the signal
measured again. In this way a plot was made of signal as a function or orientation
error. At the end of the work other objects such as slits were tested.
Experimental Equipment
The diffraction hologram of the object was carefully aligned with its spatial
filter using a 60 power microscope. The spatial filter holder incorporated an
accurately calibrated circular scale which can be read to 5 minutes of arc.
Measurements of the signal were made with the Gamma Scientific photometric micro-
scope which is linear to at least 1 percent.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C- 3- 2
Experimental Results
Every object and filter combination tested except one produced practically
identical error curves. Figure 1 is typical and shows the result for an M-48 tank
which measured 5 mm long on the object transparency.
The curves show a "plateau" near zero error and practically linear decrease
in signal beyond this plateau. Examining this curve, it is possible to say that
for objects of this general size and shape, 1/2 degree rotation error produces
uegligible signal loss, 2 degrees rotation error reduces the signal to about .80
its original strength, 4 degrees error results in a .50 signal loss and an error
or 7 to 8 degrees usually reduces the signal close to the noise level.
All of the objects tested produced such similar results that it was decided
to test two very unlikely objects. One of these was a slit 2 mm wide. The spatial
filter of this object consisted of a series of very narrow lines in a straight line.
Since these lines were not over .002" wide it was felt that a very small rotation
of the filter would certainly cause the higher order elements to be displaced from
their corresponding diffraction images and therefore a different error curve would
result. The plot turned out to be practically identical to Figure 1 which seemed
very surprising. The reason appears to be that it is possible to make a good spatial
filter of a "wide" slit only if the lower spatial frequencies are included. These
are located near the zero order and move on a short radius arm when the filter is
rotated. If the spatial filter of a "wide" slit is made to include the higher
frequencies only, it does not recognize the slit which is essentially a low frequency
object. It does recognize the edges, but this recognition is very weak. To put
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
I0
9
8
7
30
20
10
? - -
.
.
EFFECT
ON
1
OR ORIENTATION
SIGNAL
1
? TANK
1
ERROR
5mm
1
Fr.
Ill
huh
API"
bk.
El
BIM
0
3 2
Aooroved For Release 20?P2D9NIZIMEM?70A002300020010-5
0
2
3
FIGURE I
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-3-4
this conclusion in other words; the only part of the spatial filter of a wide slit
that is effective is the part near the axis and since this rotates on a short arm
the tolerance is relatively large. The higher diffraction orders are very
conspicuous in the spatial filter, but appear to be completely inoperative.
A test was also made of a "narrow" slit only .2 mm wide. As might be expected,
the first diffraction order is at a much greater distance from the zero order than
in the case of the wide slit. It would appear that this would change the rotation
tolerance, but the measured difference was insignificant. The explanation is that
wide slits produce narrow, sharp images, but narrow slits produce wide fuzzy images.
While it is true that the separation of the orders is greater in the case of the narrow
slit and therefore the images move on a longer radius arm, the images are so much
larger that the two effects practically compensate.
The only object for which we could obtain a noticeably different error function
was a random noise pattern correlated with its spatial filter. This pattern consisted
of about 40,000 irregularly shaped spots. A rotation error of 20 minutes of arc was
enough to reduce the signal to .50 its normal value. At present we regard this as an
example of a situation that would be unlikely to occur in aerial photographs.
Analytical Approach
In an earlier Section on Alignment Tolerance, a simple theory was developed to
account for the observed error curve. Most of the same ideas are applicable here.
Taking the spatial filter of a tank as an example, it consists of a number of "spots"
which are the diffraction orders of the various details. It was found that only a few of the
total number of spots were effective in the recognition process.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-3-5
The following simplifying assumptions were made:
1. The only "spots" considered were the diffraction elements caused
by the two parallel long sides of the tank.
2. The only orders considered were the 4th to 8th which were the only
ones which showed appreciable modulation efficiency.
3. All the diffraction orders are shown equally spaced. The actual
spacing is variable due to interaction of other diffraction elements.
4. All the diffraction orders are the same size. Actually the size varies
with the intensity. The spots are drawn the average size for the 5th
to 8th orders which appear to be ideally exposed and measure .200 mm
diameter.
5. The interference detail in the filter has been omitted since it is
not important in this interpretation.
Figure 2 shows the common area of the diffraction orders in the hologram of
the object and the spatial filter with various amounts of rotation error between
the two. It can be seem that the common area is approximately equal to the signal
strength.
Conclusions on Rotation Tolerance
1. The curve of rotation error as a function of relative siganl intensity
has a practically flat plateau at zero error and essentially linear
slopes.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010-5c-3-6
o
TH
7 TH
6TH
5 TH
4 TH
ZERO ORDER
ZERO ORDER ZERO ORDER
GEOMETRICAL EFFECT OF ROTATION ERROR
FIGURE 2
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-3-7
2. For the objects tested, a rotation error less than 1/2 of one degree
results in negligible signal loss. Four degrees of rotation error
reduces the signal to half its normal value.
3. The error curve is practically constant for many sizes and shapes of
objects, even long slits.
4. It appears that a satisfactory mathematical solution for any orientation
error can be derived from the geometry of the diffraction image.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-4-1
Section C-4
Magnification Tolerance of a Spatial Filter
Purpose
In aerial photographs the size of the image on the film will vary with the
altitude of the camera and the focal length of the lens. If the film is to be
scanned with a spatial filter, then some way has to be found to bring the diffraction
image of the object to the same size as the diffraction image in the filter. In the
design of an automatic system it is necessary to know the tolerance on magnification
in order to estimate the time cycle required to scan over a known degree of magnificat-
ion change.
Experimental Approach
A transparency containing an object, vehicle or character was aligned with
a spatial filter made for this object. The intensity of the recognition signal
was measured. Then the relative magnification of the object was changed and the
signal strength measured again. In this way it was possible to make a plot of
magnification difference as a function of signal strength. A number of objects
of different size and complexity were tested and the curves for all of them were
close to that shown in Figure 1.
Experimental Procedure
The optical setup for these tests is not the usual one in which the object
transparency is in collimated light because there is no simple way to vary the
relative magnification. Instead the "overlapping" system shown in Figure 2 is used.
The spatial filters for these tests also must be made on an overlapping system. As
shown in the Figure, the transparency is installed in a converging beam and not in
collimated light. If the filter was made with the object in the position shown, then
it will detect the same object in this position. The diffraction angles of the
various spatial frequencies in the object are fixed, but the size of the hologram
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
I0
9
8
70
30
20
10
Approved
For
Release
2005/05/02
.
?
? A-RDP78B0
?
770Aoctrfte1Or13F5VARIATION
.
MAGNIFICATION
I -
1
? OBJECT
1
IN
B
.
.
.
.
.
1)1 0 2 4
MAGNIFICATION ? PERCENT
Approved For Release 2005/05/02 : CNU8EI04770A002300020010-5
6
8
10
,?
SOURCE
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
OBJECT LENS
OBJECT TRANSPARENCY
SPATIAL FILTER
IMAGING LENS
oft
TRANSLATES TO VARY MAGNIFICATION
"OVERLAPPING" OPTICAL SYSTEM
FIGURE 2
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A002300026-11bIg
depends on the distance between the object and the spatial filter. Thus, with the
object as shown the diffraction image will fall exactly on the corresponding elements
of the filter; as the separation is increased the diffraction angle stays the same but
the size of the pattern gradually increases. Moved in the other direction the pattern
decreases in size. By using a calibrated scale on the transparency slide it is
possible to calculate the relative magnification knowing the focal length of the object
lens and the original position of the transparency.
One peculiarity of any diffraction system is periodicity. When the object is
twice its original size the diffraction image will be half as large and now the
second order of the object will fall on the first order of the hologram and so forth.
This is bound to happen in any system where the orders are uniformly spaced. For
complex objects such as vehicles, these secondary maxima are barely detectable because
the spatial filter contains size information on many different details and the orders
are now complex. The probability of the 4th and 8th order for example, being exactly
the same is less as the complexity of the object increases. To test this feature,
a spatial filter was made of a round hole and in this case secondary recognition
maxima occurred at exactly .5X and 2X magnification. In this simple case the intensity
of these two secondary maxima were almost half, actually 42 percent of the correct size.
The purpose of mentioning this feature is that it may arise in trying to use spatial
filters for simple shapes such as slits to detect roads in an aerial photograph.
Physical Explanation
No study was made of the physical causes of the phenomenon. It is fairly clear
from the study of the models rotation and x and y translation what happens. As the
relative magnification of the diffraction images and changes, less and less of them
overlap.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-4-5
Results
For all the objects tested, the results were practically the same. The strength
of the recognition signal fell to .80 the maximum when the magnification varied plus
or minus 3 percent, it fell to .50 when the magnification plus or minus 6 percent.
For variations of magnification less than 1 percent the loss of signal strength was
negligible.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Section C-5
Time Constants of a Scanning System
Purpose
If objects are randomly oriented in a transparency it is necessary to
conduct a rotation or orientation "search" to recognize them. This is the process
of rotating either the object transparency, the spatial filter or the image of the
transparency until the two are in alignment. In an automatic system the readout
would probably be a T.V. camera and it is necessary to perform the orientation search
in such a way that no recognition spots are missed. If we assume the T.V. scan time
to be .033 seconds, and we know the curve of recognition signal strength as a function
of orientation error, then it is possible to calculate how long an orientation search
will take.
In the same way, if objects may have a random size in the transparency it is
necessary to conduct a scale search. This is usually done by changing the effective
size of the objects by moving the transparency axially in a beam of convergent light.
The purpose of this section is to estimate the time to conduct an orientation
and scale search in an automatic system.
Experimental Data
The necessary information on the effects of variation in orientation and scale
on the signal strength were collected in two previous sections.
Discussion
It is necessary to decide what loss of signal is permissible. In the rotation
experiments it was found that 1/2 degree of error produced negligible loss, 2 degrees
of angle error reduced the signal to .8 its original strength and 4 degrees of error
reduced it to .5. At present we believe that .8 signal is a safe minimum and this
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-5-2
results in an angle error of 2 degrees. The orientation search could be conducted
either by rotation of one of the images 2 degrees per scan in a step-wise fashion,
or by continuous rotation at the rate of 2 degrees per .033 seconds. It appears that
continuous rotation is preferable from the point of view of less vibration. This
means that if we wish to keep the signal strength at least .8 that a complete scan
can be conducted in 180 x .033 or 5.04 seconds.
The figures for the scale search indicate that to obtain a signal not lower
than .8 the value for a perfect match the magnification should not vary more than
3 percent. The time to make a scale search then depends on the range of scale to be
covered. If it is plus or minus 25 percent then the scale search could be performed
in 17 steps, each one requiring one complete orientation or 17 x 5.04 or 85.7 seconds
for a complete orientation and scale search. Actually the scale variation would not
be in steps, but would consist of changing the relative magnification gradually.
Result
For the case considered, where the signal should not drop below .8 of its
maximum value and the scale is varied plus or minus 25 percent, a complete orientation
and scale scan with a T.V. readout time of .033 seconds would require 86 seconds.
Clearly, none of these values is fixed and an automatic machine would have
to have some sort of a variable programming device to adapt it to different conditions.
Conclusions
The search time for what is believed to be a typical case is about two minutes,
including time to change transparencies.
Any automatic machine must be flexible to adjust to different conditions.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A00230002001066_1
Section C-6
RECORDING MATERIALS STUDY
Purpose
The only satisfactory material for spatial filters known at the start of this
project was Kodak spectroscopic plates 649-F. The normal time to process a filter
on this material is one hour, though we believe this could be reduced to about 15
minute by the use of an automatic processing machine.
Normally, the filters are removed from the equipment for processing and have
to be returned to a position exactly on the optical axis.
The purpose of this study was to determine the feasibility of other materials
with the objects of shorter exposure time, faster processing and no removal from
the equipment.
Silver Photovanhic Materials
Spatial filters were made on spectroscopic plates type 103 and Kodak high-
contrast copy film. The resolution of these emulsions was inadequate to produce
a good record in interference detail.
Attempts were made to use Kodak High Resolution plates, but these were not
sensitive at the laser wavelength.
Several other plates and films were tested without producing usable results.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Use of Photo Plastic Recording (PPR)
This term is used
C-6-2
to denote an electro-photographic
process. One form of the recording material consists of a glass plate support
and a thin plastic film on which the image is made. The glass support is normally
2 inches square and is coated with a transparent conductive film of tin oxide. At
two edges of the support there are silver strips to contact the conductive coating.
Resistance across a typical slide is 15 ohms and the light transmission is at
least 85 percent. This support is coated with a thin film of a photoconductive
thermoplastic. The coatings used in these tests varied in thickness from 11 to
20 microns.
PPR involves the following steps:
1. In the dark, the surface of the PPR plate is exposed to a corona
discharge. This produces a surface charge in the range of 600-1000
volts. This charge can be retained for an hour or more if necessary.
2. The PPR slide is exposed to light. In the exposed portions the photo-
conductor allows the surface charge to leak through the plastic to the
conductive coating which is grounded. The unexposed areas retain their
charge.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-3
3. The PPR slide is developed by heat. This can be done in a number
of ways, but in the present experiments the plastic film was
softened by resistive heating of the conductive film. The time
required for development is .06 seconds. The effect of heating
is to soften the thermoplastic so that the electrostatic effect
of the surface charges will deform the film. On cooling these
deformations are frozen. In general, the charged areas attract
the ground plane and are depressed. The resulting image consists
of ripples in the plastic surface, a completely transparent image,
but one capable of refraction, diffraction and phase modulation.
After processing, the image may be erased by a longer heating cycle and the
slide reused. This longer cycle melts the thermoplastic and heats the photo-
conductor to the point where it becomes a conductor. This completely discharges
the slide and leaves it with a smooth surface.
Sensitivity of PPR
The #216 PPR slides are not red-sensitive. It was necessary to develop a
special formulation known as PPR 334* containing a proprietary sensitizer to be used
with laser light at 6328 Angstroms. Unlike normal PPR slides this material is not
completely transparent, but blue in color and a fairly strong absorber of red light.
The results to be reported were obtained with the very first #334 materials
prepared. The data and inferences drawn from it must therefore be viewed as
preliminary.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-4
Theory of the Use of PPR Spatial Filters
The two-beam process for making a spatial filter uses the interference
of two beams of coherent monochromatic light one of which contains diffraction
produced by the object. This complex image contains both phase and amplitude
information about the object, and normally this is recorded on silver-bearing
photographic emulsion. After processing, this filter can be used to recognize
the same object in a larger body of information. The design of the equipment is
such that both the phase and amplitude information can be recorded as light wave
amplitude or variations in density in the photographic material. Such a filter is
described as a "complex amplitude-modulating filter".
It is equally possible to record the phase and amplitude information on
a transparent deformable material such as PPR so that the resulting filter modulates
the phase of the incident light. This kind of filter is called a "complex phase-
modulating filter".
So far as we can tell from theory, there is no essential difference between
the two types of filters, there are, however, several practical differences.
Practical Differences between Am litude and Phase-Modulatin Filters
The amplitude-modulating or silver-bearing filter can exist in either a
negative (exposed portions opaque) or positive (exposed portions transparent)
filter. In some cases the performance of these two kinds of filters is quite
different. In the case of phase-modulating filters it is difficult to tell a
negative from a positive and there is no apparent difference in their performance.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-5
The amplitude filter contains opaque material (silver grains) which obstruct
part of the light incident on the filter, reducing the amount of light in all of
the transmitted images. The phase filter can be completely transparent (neglecting
surface reflection losses) and the total light in the transmitted images is essentially
equal to the incident light.
A two-beam spatial filter produces a recognition image that is off-set from the
optical axis or zero?order image. The ratio of the light in this off-set image,
compared to the light incident on the filter is termed the modulation efficiency of
the filter and is the chief factor in determining the intensity of the recognition
image. The modulation efficiency of an amplitude filter, based on reasonable
assumptions, is limited to about 16 percent. The efficiency of a phase-modulating
filter can be over 40 percent. Therefore the brightness of the recognition images
and in general the light efficiency of the whole process should be two to three
times higher with a phase filter.
Relative Speed of PPR
"Speed" in this case refers to the idea of light sensitivity for a particular
Application. A light sensitive material is said to be "fast" if it requires a short
exposure and 'slow" if it requires a long exposure.
The recommended silver photographic material for making spatial filters is
Kodak spectroscopic emulsion 649-F on either glass plates or film. There are
several problems with this material. The sensitization curve falls off rapidly
in the red so that the material is not efficiently used at the laser wavelength
of .6328A. In spite of several experiments, no better commercially *Venable material
has been found.
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-6
The most undesirable feature of 649-F is its slow speed. A typical
spatial filter on our 5-inch aperture system requires an exposure 2 to 10
minutes.
The same exposure may be obtained on red-sensitive PPR in from 5 to 20
seconds, using a charge of 33 volts per micron of thickness of the PPR coating.
It should be understood that the speed of any light-sensitive material
is difficult to define except for a specified set of conditions. The effective
speed of a photographic film can be increased through special development or
hypersensitization by a factor of ten times or more. The speed can also be
reduced by desensitization and underdevelopment. In exactly the same way, the
speed of PPR is not a fixed quantity, but depends on the charging stress and
can be altered by changing the development time. It is possible to obtain
considerably higher speed on PPR than that reported above by using a higher
surface charge, but such slides were "frosty" or noisy and make poor spatial
filters. (There have been substantial improvements.)
The conclusion on speed is that red-sensitive PPR is at least 50 to 100 times
faster than the only satisfactory silver material presently available. This is an
advantage in the case of poorly transmitting objects where the exposure can be
reduced from an hour to 40 seconds.
Surface Quality of PPR Slides
The two separate optical systems used to make and use spatial filters are
close to "perfect" systems which means that they are performing close to the
limit to which light waves can be focused and controlled. It can be predicted
from theory or easily confirmed by experiment that relatively minor departures
from this perfection will cause a major reduction in the recognition signal-to-
noise ratio.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-7
These imperfections are easily introduced in the system by installing a
glass plate that is not uniform in thickness or refractive index or which has
obvious light-scattering defects on its surfaces such as scratches or roughness.
This problem is common to silver-bearing and PPR materials, with one important
difference; means has been found for compensating for the defects in silver-bearing
amplitude materials, but no means of compensation is known in the case of PPR phase-
modulating spatial filters.
If the glass plates on which silver filters are made are not flat or uniform
in thickness they introduce deviations in the diffracted rays. It should be
recalled that the diffraction angles realized in this system are small, only a few
minutes of arc at most, and into this few minutes is crowded a tremendous amount of
detail. The size of this detail is more or less determined by the diameter of the
zero order diffraction image of the source, and in a typical instrument will be
about two seconds of arc. As a result of this, any glass plate that randomly
deviates or scatters an appreciable portion of the rays by two seconds of angle
or more will appreciably degrade the system. In practical terms this means that
a plate of clear glass which appears perfect to the eye may, when added to the system
cause it to stop working. Most of the glass plates on which photographic emulsions
are coated are not of a quality good enough to place in one of these optical systems.
As result, liquid gates have been developed to "compensate" this effect. A liquid gate is a
transparent cell with two parallel windows made of tested high-quality glass. The cell
is filled with a liquid having the same refractive index as the glass and when a low-
quality glass plate is placed in the cell the effect is to smooth out the irregular-
ities in the glass surface. In effect, the cell windows, liquid and photographic
plate all have the same index and so become optically a solid piece with two near-
perfect surfaces, the cell windows.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020(0_03-5
We have tested several of the glass plates on which PPR is normally coated
and all of the plates tested to date have been adequate for spatial filters.
This does not mean they are "perfect", they usually show detectable degradation
of the diffraction image, but this appears to be intermediate in amount between
normal glass photographic plates which are mostly usable and "Microflat" glass
which is the best material we can obtain and is virtually perfect. At present
we believe that the surface quality of the glass plate or support is not a problem.
Such defects as we have noted in PPR slides were attributed to three causes:
1. Irregularities in the thickness of the coating, especially at the
edges. This can be avoided by using only the center and masking
the edges.
2. Dust embedded in the surface of the slide. This is rarely present
in a new slide, but increases as the slide is used as described in a
paragraph titled, "Reusability".
3. Overall roughness of the PPR surface which either cannot be erased
or which returns on redevelopment.
At the present time these slides are handmade and a dust-free environment
is not available for their production. It is expected that future improvements
in coating technique will nearly eliminate these defects.
The conclusion surface quality is that selected PPR plates are adequate for
any spatial filters that have been attempted. The selection process is simple and
fast. About 75 percent of the first batch of slides was usable.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-9
Resolution of PPR
One of the requirements for material for spatial filters is high resolution
in order to be able to record fine interference detail.
The resolution of spectroscopic emulsion 649-F is so high that it cannot be
measured optically. This means, that judged from measurements of the grain size
and attempts to form very fine optical patterns with microscope lenses, the
resolution is probably in excess of 1000 lines/mm. As mentioned earlier, this
is the only completely satisfactory silver-bearing photographic material yet
found.
The resolution of Kodak high contrast copy film when given ideal development
is about 250 lines/mm. We have made many spatial filters of identical objects
on both these materials and there is no question that the spectroscopic film
produces a spatial filter with a higher modulation efficiency and one which
results in a crisper reconstructed image.
When emulsions showing less than 150 lines/mm are used, the resulting
spatial filters show very low efficiency, and have proven to be useless.
Spatial filters made on red-sensitive PPR appear to be equal to those made
on spectroscopic emulsion in every test we have been able to make and also in
general appearance of the detail when examined under a microscope at high power.
The conclusion at present is that PPR has adequate resolution and this must be
greater than 250 1/mm and probably is in the region of 1000 1/mm.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-10
PPR Equipment
The special equipment for handling PPR consists of the following
assemblies:
1. Charging and development station.
2. Charging head and power supply.
3. Development power supply and timer.
4. Mechanical traversing device.
5. Charge measuring head and meter.
6. Development microscope.
7. Slide evaluator.
1. Charging and Development Station
This equipment is essentially a holder for the PPR slide and is built on
an optical bench carriage. The PPR slide is held in a vertical position which
is not the normal practice, but which seems to have given no trouble.
Spring-finger contacts are provided to make electrical contact to the
slide. The slide holder is designed to position the slide by its front surface
so that variation in slide thickness will not affect the focus.
This station also supports the mechanical traversing device with its
charging and measuring heads and the charge measuring meter.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-11
2. Charging Head and Power Supply
The charging head consists of a fine tungsten wire surrounded on three
sides by shields. When 5,000 to 10,000 volts is applied to this small wire it
produces a corona discharge which charges the dielectric surface of the slide.
A uniform charge is produced by traversing this head across the slide. Two
passes are normally made.
The power supply consists of a calibrated variable transformer and a
solid state 10 KVDC power pack. Normally voltages between 5,000 and 8,500 volts
are used to charge the wire in the charging head. This will produce a surface
charge on the slide between 400 and 800 volts. The chief control on surface
charge is the charge on the head which can be varied by the variable transformer.
3. Development Power Supply and Timer
The power used to heat the PPR slide for development or erasure comes from
a variable transformer. The total power used is 480 watts, the voltage being
adjusted according to the slide resistance.
The development or erasure cycle is timed by a cycle timer developed by
this is a solid state device which can be set to pass any number
of full cycles of 60 cycle current from 1 to 24. A feature of the device is that
It cuts on and off at the zero voltage point of the cycle so that not only is the
number of cycles exact, but each one is exactly a full cycle.
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-12
In the experiments described, the recommended development times were as
follows:
Development - 7 cycles
Erasure - 24 cycles
The development circuit also contains a 1 KW isolation transformer. This
is necessitated by the fact that the conductive coating on the PPR slide is used
for two purposes; to heat the slide and also to act as a ground plane for the
surface charge. For this latter purpose it is absolutely necessary that one side
of the slide be grounded to a real, not a nominal ground. In fact, one side of
the high-voltage charging circuit, the electrostatic voltmeter and the corona shield
all have to be directly connected to this same ground. Since it is unsafe to ground
one side of an alternating current power line, the isolation transformer is used to
separate this circuit.
4. Mechanical Traversing Device
This assembly consists of a pair of accurate rails, a motor-driven carriage
supporting the charging head and charge measuring head. It is essential that these
units maintain a constant distance from the slide and that the charging head move
at a constant velocity. Minor variations in distance will produce major variations
in the charge or the apparent measured charge.
5. Charge Measuring Head and Meter
The measuring head is a unit developed at this laboratory and is standard on
all PPR equipment. For proper operation it is positioned .040" from the
sLide surface.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-13
The meter is a Keithley electrostatic voltmeter, model 200B. It is
installed on the charging and development station because the lead length from
measuring head to meter should be as short as possible and not over three feet.
When the surface charge on a slide is about 600 volts the charge induced in the
measuring head is 0.5 volt. The measuring head has to be calibrated so that
the meter readings can be converted to actual surface charge.
There are four quite different voltages in the charging system, all of
which have to be correlated:
a. The voltage shown on the variable transformer input to the
high voltage power supply. This is the primary control of
charging voltage and varies between 50 and 85 volts.
b. The voltage on the charging head which varies between 5,000
and 8,500 volts.
c. The actual charge on the 'slide. This varies between 400 and
800 volts.
d. The actual voltage on the measuring head and indicated by the
meter. This varies between .4 and .8 volts.
6. Development Microscope
Normally, PPR images are developed by inspection. An instrument for red-
sensitive PPR has presented several serious problems and no solution was found
in time for inclusion in this project.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-14
7. Slide Evaluator
It is very difficult to see the diffraction image of a spatial filter on a
PPR slide. This makes it impossible to look at the filter and determine whether
or not it is properly exposed and developed, contains defects, etc.
The fine detail of the filter can be seen and photographed in a phase-
contrast microscope. Since these instruments are not available at less than
100X this means that only a small area of the filter can be seen. A typical
spatial filter is 20 mm in diameter, while the field of view of a 100X micro-
scope is only about 2 mm.
One of the desirable measurements in evaluating spatial filters is the
number of orders of diffraction recorded. A crude reflected-light schlieren device
was made for this purpose and proved to be indispensible.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-15
DESCRIPTION OF PHOTOS OF PPR EQUIPMENT
Figure 1 -
General view of complete PPR equipment on optical bench. PPR slide
is shown in position on optical axis. Carriage supporting charging head is
in extreme left position. Meter at top reads surface charge.
Under the bench the charging power supply has the high voltage control
and motor traversing switch. The cycle timer, regulates the length of the
development cycle and is set to 7 cycles. The development power supply at left
contains the variable and isolation transformers.
Figure 2 -
Closeup of the equipment surrounding the PPR slide. The slide surface is
located by the Teflon guides on both sides. Electrical contact is made by spring
finger contacts at top and bottom.
The carriage rides on two round rails at top and bottom. The reversible
traversing motor is mostly hidden by the carriage) a pinion on the shaft engages
the rack above the top rail. Attached to the carriage and moving with it are the
charging head with high voltage wires leading to it and grounding shields surrounding
it. To the left is the charge measuring head which can be positioned over the center
of the slide when the carriage is traversed to the right.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-16
! Charging Power Supply
Cycle Timer
Development Power Supply
FIGURE 1 (ATL Photo 842401)
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Electrical Contacts to Slide
FIGURE 2 (ATL Photo 8424021
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-18
Making a PPR Spatial Filter
The process of making these filters has not been automated to the point
where it is a matter of pressing a button. It appears to be entirely feasible
to automate the process, but this was not considered in our experimental setup.
The first three steps are common to the process for making either a PPR
or silver photographic filter, they are mentioned only to give an idea of the
preparation required.
1. An object is selected and made into a transparency.
2. The object is placed in the sample beam and the reference beam
is attenuated to a value which produces optimum interference in
the desired diffraction order. This decision influences the
selectivity of the final spatial filter.
3. The exposure is calculated to produce a properly exposed filter.
Failure to achieve this results in a filter with low modulation
efficiency.
4. A PPR slide is installed at the focus of the diffraction image.
5. The PPR slide is "cleared" if necessary. PPR coatings are hygroscopic
and absorb moisture which -produces a rough surface when the film is
developed. To remove this moisture the slide is "cleared". The
clearing process is erasure, except that in general, no image is
present. The present clearing cycle is 21 cycles at 480 watts.
To obtain this, the resistance value of the slide is converted by
a chart to a voltage value to be applied to the development circuit
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-19
and this value is manually set on the variable transformer and the
clearing or erasure cycle completed. If the slide
within the previous 8 hours then it can
cleared and this step is not required.
6. The PPR slide, if cleared, is cooled to
has been erased
be assumed that it has been
room temperature. The time
required depends on the heat sink available.
7. The PPR slide is charged. In
safelight, the charge head is
while applying a high voltage
total darkness, or using a tested
traversed across the slide twice
to the head.
The voltage is regulated
by a calibrated variable transformer and is adjusted by experience
to produce a surface charge of 33 volts per micron of thickness in
the PPR coating. The thickness of the coating can be obtained from
the slide container and the transformer is manually adjusted to the
proper value.
8. The charge on the PPR is measured. This step should not be necessary
in routine work, but it is
measuring head is attached
To make a measurement, the
stray atmospheric
static voltmeter,
the slide and the
essential at the present time. The charge
to the same carriage as the charging head.
measuring head is first discharged of any
charge by pressing the shorting button on the electro-
then the measuring head is moved over the center of
charge read. It is helpful to have a safe-lighted
meter. The meter reading is converted by a chart to give the actual slide
charge which should be close to that calculated in the last step. If the
charge is too low it may be increased by extra charging passes or by
increasing the charge on the head and making extra passes. It may be desirable
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-20
at this time to observe the meter for at least 15 seconds to determine
if the charge is leaking. A decreasing charge can result from light
leakage, a poor slide, or a slide that is not fully cooled. A warm
slide will not hold a charge.
9. The charge measuring head is moved out of the way.
10. The PPR slide is exposed. At present this is done manually
using a stop watch.
11. The PPR slide is developed. The voltage in the developing
circuit is manually set at described under 5 above, and the timer
set to recommended development time which is 7 cycles and the
"start" button pressed.
This completes the PPR spatial filter. As can be seen, the operation at
present involves a number of hand motions, decisions and corrections and may
take about three minutes of elapsed time. This could be reduced to about 20
seconds plus exposure time.
Modulation Efficiency
I. Theoretical
The silver photographic process produces an "amplitude" spatial filter.
The theoretical maximum modulation efficiency of such a filter is 16.4 percent.
This calculation was based on a number of assumptions that are not necessarily
correct:
a. The opaque lines are totally opaque.
b. The transparent lines are completely transparent.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-21
c. The width of the opaque and transparent lines is equal.
In a photographic grating it is extremely difficult to meet these conditions
exactly, and actual filters have shown measured modulation efficiencies between 2
and 10 percent. Actually, the situation is more complex, because there is usually
a surface relief image on the silver film which acts as a phase grating. Thus,
the usual silver photograph spatial filter is a combination phase and amplitude
filter in which the opacity of the amplitude image attenuates the efficiency of
the phase image. As a result, silver filters can have efficiencies higher than
the theoretical 16.4 percent and 20 percent has been measured.
A completely transparent phase-grating such as produced on PPR can theoretically
show a modulatiOn efficiency over 40 percent, this being one of the advantages of
this material. Actual samples of PPR tested on the two modulation efficiency
instruments consistently show values in the neighborhood of 40 percent if correctly
exposed and developed.
2. Actual Results on Spatial Filters
The first phase-modulating spatial filters were made by contact printing a
silver filter on blue-sensitive PPR. These filters were "noisy" due to the
mechanics of the contact printing process, but they showed modulation efficiencies
of about 35 percent which was higher than we have ever obtained or can expect to
obtain with silver filters.
The modulation efficiency of red-sensitive PPR filters has measured about
2 percent. The reason for this is that the 334 PPR formulation contains a red-
absorbing sensitizer, which in typical slides has a transmission for laser 1%ght
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-22
of only 5 percent. This value can be obtained either from the spectrophotometric
curve or by measuring the optical density of the slide in red light. If such a
slide had an inherent modulation efficiency of 40 percent and a transmission for
red light of 5 percent the measured modulation efficiency in the recognition setup
would be only 2 percent which is the actual value obtained.
An attempt was made to increase the red light transmission of the slides by
reducing the sensitizer content, but no images were obtainable on these slides.
Apparently, there is a minimum sensitizer content below which the material fails
to form a conducting chain through the dielectric film.
The conclusion on efficiency is that the inherent modulation efficiency of
PPR is about twice as high as silver-bearing photographic film. This has been
measured on PPR containing no absorbing material. The code 334 red-sensitive
formulation contains a red-absorbing sensitizer which transmits only about 5 percent
of the incident light resulting in a low efficiency. This is the first red-
sensitive formulation tested. There is every reason to believe an improved
material can be found.
Reusability of PPR
Theoretically, PPR is reusable indefinitely. At the end of considerable use
the image tends to become "sticky" (difficult to erase);the modulation depth is
less (lower efficiency). It is possible to predict that continued development
and erasure heat cycles will evaporate the more volatile portions of the film
and result in a thinner, denser coating. It is also possible that continued
heating may degrade the sensitizer and cause a loss of sensitivity.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-23
In practice, none of these effects has been of any importance because
when the images are made and developed in air they attract dust which becomes
part of the surface during the erasure cycle and after several cycles the slides
become too dusty or "noisy" to use.
In our experiments, no attempt was made to exclude dust or determine the
maximum reusability of a PPR slide. We found that a slide could be used about
15 times before the "noise" became objectionable. Several circumstances should
be pointed out. First, the charging cycle places a strong static charge on the
dielectric surface of the slide. This appears to attract the nearby dust, also
the corona discharge probably accelerates the dust in the air between the
charging head and the slide and drives it onto the slide with considerable
force. Secondly, these slides are removed from the equipment for other tests
and have ample opportunity to collect laboratory dust, more than they would if
kept in the equipment for routine use.
It has been suggested that if the area around the charge and development
station were enclosed and supplied with relatively dust-free air that the
reusability of the PPR slides might be extended to a 100 or more uses.
PPR Experiments
A typical image of a model tank was selected as an object and spatial
filters made on PPR and the best silver photographic material available, Kodak
spectroscopic plates 649-F on Microflat glass. In each case the exposure was
adjusted to give an image which showed the same amount of detail when examined
under a suitable microscope. In the case of silver-bearing filters the maximum
density was recorded and for PPR filters the modulation depth was measured.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-6-24
The silver and PPR spatial filters were then used to recognize similar
objects in other transparencies. The tolerance of alignment, the recognition
signal strength and the signal-to-noise ratio were then measured for both cases.
So far as could be told, the properties of a good PPR spatial filter are no
different from those of a good silver-bearing filter.
The only difference between results obtained with the two kinds of filters
was that the recognition signal strength was about one-tenth as high when PPR
filters were used. This is contrary to the predicted higher efficiency of PPR
filters and is due to the absorption of red light in the present red-sensitive
PPR material. This is described separately under "Modulation Efficiency".
We have calculated that if the PPR slides were colorless, their recognition
signals would have been 2 to 3 times higher than that of the silver filters.
Results
1. The signal-to-noise ratio of the recognition signal is equal
for equivalent silver and PPR spatial filters.
2. The absolute signal is stronger with non-absorbing PPR than the
test silver filters. Red-sensitive PPR material has low trans-
mission for red light and the signal is attenuated by this
absorption.
3. PPR filters can be processed in much less time than silver filters.
4. PPR material has adequate resolution.
5. Red-sensitive PPR requires a much shorter exposure time than
satisfactory silver material.
6. PPR material is reusable.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP781304770A0023000206E1-17-51
Section C-7
FEASIBILITY OF REAL-TIME FILTER GENERATION
In a strict sense "real-time" means instantaneous, but in connection with
systems development it has come to mean any short time which is comparable to
other necessary short delays in the process. It implies that the step under
consideration is not the time bottleneck in the processing cycle. If we are
willing to accept this more liberal interpretation of the term, then Photo-
plastic Recording (PPR) has been demonstrated as a successful means for real-
time filter generation.
We have looked at this two ways:
I. The total elapsed time to make a silver-bearing filter on
649 material was two minutes exposure plus one hour processing.
The total elapsed time to make a PPR filter of the same subject
was 10 seconds exposure plus twenty seconds charging time plus
a fraction of a second development. From this point of view, the
PPR filter is finished before the silver filter was even exposed.
2. If we regard the exposure time as a separate item and count as the
time for filter generation only the processing interval, then the
present PPR processing time could easily be as short as 20 seconds,
or less if the slides were pre-charged. Since this period is short
compared to the exposure time, it can be regarded as not seriously
delaying the process.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A00230002001 %-58_
Section C-8
Television Readout of Recognition Signals
Introduction
Some sort of electrical readout is a necessity. Due to the high noise
background of the image as a whole, it is not possible to use a simple photocell
to detect the recognition signals. These recognition spots are bright, but they
are so small that they add very little light to the total integrated light in
the image. The only readout that appears feasible at present is a closed-
circuit TV chain in which the camera scans the image and large increases in the
video signal will be produced when the scan pattern crosses a recognition spot.
In its simplest form, this signal would trip a counter when it was above a
specified noise level. The TV chain used, without the counters is shown in
Fig. 1.
In practice something more sophisticated is required. There are two
reasons for this. First, tests to date indicate that a recognition spot appears
not in one scan line only, but in two or more. The system we have used is non-
interlaced to avoid the problems of successive lines occurring in different frames.
Under ideal conditions, when the center of a recognition spot is coincident with
the center of an individual horizontal scan line, a strong video signals is generated
as the scan line passes through the spot. Successively weaker signals appear in the
one or two lines immediately above and below the center of the spot. Alternatively,
if the center of the recognition spot falls between scan lines, two pulses of more
or less equal amplitude occur in the two adjacent lines, again with weaker pulses
above and below. Throughout our tests, close inspection of the monitor screen as
well as CRO traces of individual lines of video signal showed pulses on between one
and five successive lines depending upon spot intensity and position. Fig. 2a shows
the CRO traces of six successive scan lines. The recognition shows as a spike on the
3rd, 4th, 5th and 6th lines.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
FIGURE 1 T.V. Readout System
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-8-3
2a CRO Trace of 6 Line Scans
2b TV Monitor Showing Two Recognition Images
(Greatly Enlarged)
Approved For Release 2005/05/02 : qlfkgMD7281304770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP781304770A0023000200q054
The first requirement, therefore, is a logic circuit to prevent multiple
counting of a single recognition image which may appear in two or more lines.
One way to do this would be to put the first signal in storage with a note of
its horizontal position in the scan line. Successive signals in the next three
scan lines are then placed in the same storage if they have the same horizontal
position. At the end of four scans the memory is cleared and the stored signals
read out as one count. This same logic circuit can also be used to provide a
degree of immunity against false alarms by ignoring signals that occur in one line
only.
The second requirement is a logic circuit to prevent multiple counting of
a single recognition image which may appear in several successive frames during
the orientation and scale search. Tests have shown that during orientation scan,
for example, a recognition spot will appear at low, but detectible intensity as
the filter becomes aligned with the target, in the next frame where the alignment
is assumed to be optimum the signal is maximum intensity and in the third frame
a weak signal is received. If we assume the object is stationary and the filter
rotates, then these three recognition signals will all occur in the same place.
To prevent reading the several recognition signals for one target as three or more
targets it will be necessary to store the signals with information on their
horizontal and vertical position for several frames. All signals with a position
will go in the same storage and then periodically the storage will be cleared and
read as one signal.
Tests Performed Using Television Chain
The image produced by the recognition system was viewed with an Image Orthicon
Camera equipped with a tube having a highly red-sensitive (S-20) photosurface. This
image tube has previously been found to perform quite well at the 6943 A ruby laser
wavelength when used for nighttime photography. Its sensitivity for the gas laser
wavelength of 6328 A should be even higher.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-8-5
Two recognition spots were visible in the image. The maximum intensities
for the two spots which were somewhat different occurred at two different angular
positions. Fig. 2b is a photograph of the TV monitor screen when the object
angular position was halfway between the two maxima. The CRO traces shown in
Fig. 3a and b show the shape of the video pulse produced as the I-0 target is
scanned through the spot. The amplitude of the recognition spot relative to
background in 3a can be seen to be about 3.5 to 1. Since, in general, there is
no information contained in the background signal, the video amplifier system can
be adjusted so that almost all but the recognition spot signal is below the "black"
level giving an effective signal-to-noise ratio of 20 to 1 as seen in 3b where the
noise has been suppressed.
As mentioned above, close inspection of the monitor' screen and the CRO
traces show video pulses in from 1 to 5 scan lines depending on recognition spot
intensity and position. Feeding the video signal into an electronic scaler using
only amplitude discrimination indicated a count of from 2 to 4 per frame scan for
the more intense spot and from 2 to 3 per frame for the less intense one. On the
basis of the pulse count, the peak intensity points were found to be at an angular
separation of about 3 to 4 degrees. The overall angular span in which the more
intense spot was detected was 12 degrees while the less intense one was detected
over a nine degree range. Outside the limits of detection the pulse count dropped
quickly (less than 1 degree of rotation) from 2 per frame to zero.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
C-8-6
3a CRO Trace of Single Line Scan Expanded
3b CRO Trace of Single Line Scan, Noise Suppressed
Approved For Release 2005/9Wr?eCtA-RDP78B04770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP781304770A00230002&5436-i
Section C
Conclusions
1. It appears feasible to make an automatic device to change spatial filters.
2. The curves of recognition signal strength as a function of orientation or
magnification error are very favorable for machine recognition and provide
data for calculating the time to scan a frame.
3. The time to scan a frame depends on the range of magnification to be
covered. If this is assumed to be 2:1, then the time to scan one frame
is less two minutes.
4. It is possible to make spatial filters by Photoplastic recording in a very
short time and these filters perform practically the same as silver
photographic filters.
5. The Photoplastic process is the only means we know of for real-time
filter generation.
6. Television readout of the signals has been satisfactory, but will require
some special circuits for automatic use.
7. The general conclusion is that answers or partial solutions have been
found to several of the problems of automating the system and adapting
it to operate in real-time. Considerably more work on automation remains
to be done, but no serious problems have been uncovered.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010/.51-1
Section D
Problems of the Photograph
This section was concerned with the effects of image variables on the
detectability of targets. Typical problems were the effects of camouflage
or low-contrast on the recognition system. These tests were all made using
simulated aerial photographs and it is planned at some future time to check
as many as possible of them on actual aerial camera film. The following topics
were considered:
D-1 Properties of the Simulated Aerial Photographs
D-2 Effect of Target Size
D-3 Effect of Multiple Identical Targets
D-4 Multiple-Object Filter
D-5 Effects of Obscuration
D-6 Effect of Target Aspect .
D-7 Effect of Shadows
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release Ngigto3): FIA-RDP781304770A002300020D0-110.-i
Simulated Aerial Photographs
Purpose
The object of one part of this program was to find the effect of image
variables such as the angle of illumination of the target. It would not be
possible to measure the effect of this variable if the photographs to be tested
also varied in density, contrast, sharpness, graininess or any of the other usual
image parameters. For this reason it was decided to make these tests from simulated
aerial photographs taken of models in our laboratory rather than try to find suitable
frames on actual aerial films.
Procedure
There was no attempt to make the simulated aerial photographs resemble actual
aerial films by building a set with trees, houses, roads and all the other usual
items. In our opinion this would have been wasted effort. The only features that
had to be simulated were the target, or "several targets and the "noise" background.
In order to obtain a uniform signal from the target and noise background in
a series of pictures it was necessary to control the following variables:
1. Contrast. This was adjusted to a contrast ratio of 1.5 which we believe
is normal for aerial photographs.
2. Sharpness. All photographs were adjusted to show a resolution of
about 150 lines per mm which we believed to be a high, but not unrealistic value.
3. Density. The photographs were adjusted to show approximately the same
maximum density in any series, this was usually about maximum density 1.7.
4. Noise. The noise background was simulated by using fine sand, which
when properly illuminated produced a noise reading very close to that of actual
aerial photographs.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010351-3
5. Graininess. The photographs were taken on very fine-grained spectro-
scopic plates, Kodak 649-F, but the graininess aerial film was simulated by the
sand grains of the background.
The photographs were made on Kodak microflat glass so that the quality
of the support would be as uniform as possible and there would be no necessity
for liquid gates.
It proved to be more difficult than anticipated to make a set of photo-
graphs showing only one variable. For example, it was nearly impossible to make
a set of photographs showing variable angle of lighting and still showing constant
density. Also as the lighting angle changed the noise level of the background
changed, but this is probably a natural condition.
Conclusions
The relative signal strength and ndise background values obtained from the
simulated aerial photographs were very close to the values obtained on original
negative aerial films. We believe that the simulation was realistic and served
its purpose in providing data for the picture variables studied.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Section D-2
Packing Density or Effect of Object Size
Purpose
The purpose of these tests was to determine the effect of the size which an
object appears in the object transparency as it affects itci recognition.
Discussion
25X1 It has been shown
and confirmed by many tests that the ideal
size for the object to be recognized is about one-four hundredth the focal length
of the focusing lens. The practical explanation is that an object this size makes
the best diffraction image. If the object is very much larger than the ideal size
and approaches the size of the field of view, then its diffraction angle will be so
small that the resulting hologram will be crowded and difficult to photograph. In
general the spatial filters made from these objects will be small and inefficient.
If the object is considerably smaller than the ideal size, the diffraction
angles will be so large that the higher diffraction orders will fall outside the
spatial filter area. The field of view of the lenses used in the system is small,
and therefore the area which can be photographed and which will produce a good filter
is also small. If an object is so small that the third and higher orders fall outside
of the spatial filter area, then it will be impossible to make a high order spatial
filter of this object.
Experimental Program
A series of 4 x 5 inch negatives was made of four model tanks on a "noisy"
background resembling a ploughed field. The tank images on different negatives varied
between 10 mm and 1 mm in length. In the classical recognition system in which the
object transparency is in collimated light the object f value can be found from the
familiar equation:
? focal length object lens
f
length of object
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
1)-2-2
In the overlapping system where the object is placed in a converging beam
in order to vary its effective size, the object f value is calculated as follows:
f = distance object to lens
length of object
The object f value for the series of negatives tested is given below:
Ima,Le Size Object f Value
10 mm 123
5 246
3 410
615
1230
It was the intention to make the five spatial filters as much alike as
possible, but this did seem to be entirely practical. The observed differences
between the filters was:
Diffraction orders
Most efficient orders
10 mm
Object Size
2 mm
3 mm
33
10-12
23
4-8
19
7-15
14
4-7
7
2-5
Dia. spatial filter mm
12
16
24
24
18
Dia. diffraction orders mm
.10
.20
.3
.5
.8
Interference lines per order
10
20
33
50
100
Relative mod. eff.
.80
.80
1.0
.80
.80
Attenuaticn ret. beam
none
.20
.10
.02
.006
Exp. time 243 film sec.
.2
.5
1.0
3.0
16
Reconstruction detail
v. good
good
good
fair
poor
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-2-3
It can be seen from the Table that the five filters are not comparable,
nor does it appear practical to make them so. The filter for the 10 mm object
was most efficient in the 10-12th orders. If the 1 mm filter had extended to the
12th order it would have been larger than the camera we use to make filters or larger
than the good image produced by the lenses. At the time the filter for the 10 mm
object was made we had no way to make one showing a lower order of interference. The
object is large, transmits a large amount of light and the diffracted orders are small
and intense. This means there is a large amount of light in the higher orders. To
produce a low-order filter would require attenuating the sample beam which is not
recommended. The only other solution is to increase the intensity of the reference
beam and at the time there was no way to do this; therefore the 10-12th order filter
was the lowest we could make.
As the size of the object decreases 'the amount of light it transmits decreases
and so the exposure time to make a filter increases. This is not a linear relation
for two reasons:
a. It is usually necessary to attenuate the reference beam for small objects.
This reduces the total light available at the filter plane.
b. As the size of the object decreases the diameter of the diffraction
"spots" increases. This means that the light is spread over a larger
area and the exposure is relatively longer.
Experimental Results
(a) 10 mm Images
Filter 1460 tested with two different transparencies, one at a moderately
high contrast, the other at a fairly low contrast. The results were as follows:
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-2-4
High Contrast
Low Contrast
Signal Strength
Tank A
90
95
B
96
110
C
46
85
D
85
90
Average Noise
.5
6
Max. Noise
10
14
Min. Signal/Max.
Noise
4.6:1
6:1
The higher contrast transparency did not improve the results in this test.
(b) 5 mm Images
Signal Strength
Tank A 110
100
96
100
Average Noise .03
Max. Noise .08
Min. Signal/Max. Noise 12:1
(c) 3 Tran Images
Signal Strength
Tank A 90
120
160
140
Average Noise 3
Max. Noise 5
Min. Signal/Max. Noise 18:1
(d) 2 mm Images
Signal .Strength
Tank A 40
50
40
D 42
Average Noise 15
Max. Noise 25
Min. Signal/Max. Noise 1.6:1
(e) 1 mm Images
Signal Strength
Tank A 10
12
10
9
Max. Noise 7
Min. Signal/Max. Noise 1.3:1
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-2-5
Discussion of Results
These examples show the best signal-to-noise ratio for objects having a size
close to the ideal, or for 5 and 3 mm objects in this case. Larger and smaller
objects gave poorer signals. Of course, the filters of the different size objects
were not comparable, nor did it appear practical to make them so.
We feel that at least part of the trouble was the filters. The poor result
with the large 10 mm object was probably due partly to the use of a very high
order filter which is quite dense at the center. The poor results with the small
1 and 2 mm objects was due partly to the low order of the filter and partly to the
small size of the object. Only light transmitted by the object can contribute to
the recognition spot. As the object becomes smaller there is less light available
and so the signal strength does down. The noise, however, is approximately constant,
so that the signal-to-noise ratio decreases as the object size decreases.
A final test was made to determine if the decrease in signal observed in the
case of the small objects was due primarily to the filter or the object size. In
this test, the filter for the 3 mm object which had given good results with the
appropriate transparency was compared to the transparency with the 2 mm objects at
the proper magnification. The results were slightly better than when the 2 mm filter
was used, but not enough to make any difference. Our interpretation is that the
primary difficulty is the small size of the object and not the filter.
Conclusions
The optimum signal-to-noise ratio in continuous tone pictures was obtained for
objects close to the recommended size (one-four hundredth the focal length of the
imaging lens). Objects 2 mm or less gave poor recognition. This appears to be a
limitation of the process though it is not necessarily applicable to high contrast
objects such as microfilm copies of documents.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A00230002001_0-5
D-z-6
Smaller objects could be processed in a different setup using shorter
focal length lenses, but in this case the entire 4 x 5 inch transparency
could not be covered at one time.
In the theoretical section A-6 it was shown that objects as small as
.3 mm should be detectable. In the theoretical analysis no assumption was
made as to the design of the optical system or the focal length of the lenses,
it was assumed that the lens focal lengths were completely flexible which they
are not in practice.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A0023000200101553_1
Section D-3
Effect of Multiple Identical Targets
Object
The purpose of this test was to determine if the presence of several
identical targets in the same field of view would decrease the liklihood of their
detection.
Discussion
The theory of the process indicates that the number of targets should have
no effect and experiments with alphanumerics tended to prove this. It is less
easy to visualize the diffraction effects due more complex objects particularly
when they arranged in uniform and slightly random patterns. For example, if two
identical vehicles were oriented so that the angle between them was about 5 degrees,
the two differently oriented diffraction images would both be symmetrical about the
optical axis and the various diffraction orders would overlap and interfere. It
can be predicted that this phase interference will produce a phase shift and this
will vary the strength of the recognition signal through this particular order
of the filter.
A similar but different problem arises when a number of vehicles are parked
in a close array as in a motor pool or parking lot. In this case the edges or
ends of the vehicles are often close together so that their outline cannot be
directly seen or is in very low contrast, as in the case of a group of black
automobiles parked closely together. From a study of spatial filters we have made,
it appears that the outline is one of the most important features in recognizing
an object such as a vehicle. If this outline is lost by being merged with other
vehicles it is easy to see that the recognition may be lost or reduced.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP781304770A002300020014:1-f_2
Experimental Procedure
The test was divided into two parts; random orientation effects and close-
packing effects.
I. Random Orientation
For this test a simulated aerial photograph was used showing eleven model
tanks of the same type distributed over the picture area. The orientation angle
between these varied from one-half to ten degrees. While these vehicles were all
initially "identical" in that they were made in the same mold, all of them had been
sandblasted to a dull, dusty finish and each was dirty in a different way. The
turrets of the tanks and other minor parts were differently orientated on different
models. Also, because the photograph was not exactly vertical and the field angle
was about 30 degrees all of the tank images were not seem in the same perspective nor
did they have the same shadows around them. Therefore, the objects were "identical",
but their images on the negative were not. This is a situation similar to real life
where vehicles leave the factory in identical condition, but soon acquire an individual-
ity of their own.
Results of Random Orientation Test
In most cases the signal measured for each vehicle was identical regardless of
its position in the picture or its orientation in relation to other vehicles. The
only exceptions were vehicles orientated at small angles, of about two degrees to
each other. We interpret this as due to phase interference between the diffraction
images. This angle is of two degrees is correct to cause overlap of the most important
diffraction orders in the apparatus used. In this case one of the two recognition
signals was reduced to about 70 percent of its normal value.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D- 3- 3
Larger and smaller angles produced negligible loss of signal.
Numerical Results
Effect of Relative Orientation Angle
Angle Between Vehicles
Signal Strength
Vehicle A
Vehicle B
10
100
97
8
100
96
6
100
98
4
100
96
3
100
95
2
100
72
1
100
98
1/2
100
98
0
100
98
2. Close-Packing Effects
For this test a simulated aerial phOtograph was made showing vehicles of the
same types in groups of two, threes and sixes. Recognition signals were measured
for all the images.
Results of Close-Packing Tests
The outer vehicles in a close-packed group give normal or nearly normal
signals while the inner vehicles tend to give weak signals which may be as low as
47 percent of the normal values. This is interpreted as a loss of outline of the
inner vehicles. The outline is a major contributor to the total signal.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-3-4
Numerical Results
Effect of Close Packing
Signal Strength
Array Vehicle A Vehicle B
Two end to end 100 96
Two side by side 100 94
Three end to end, A 100 92
is end, B Center
Three side by side, A 100 54
is side, B Center
Six in a square, A 100 47
is outside, B Center
Conclusions
1. Multiple identical targets in the same photograph can result in some
loss of detection signal for some of these targets. If these vehicles
are oriented at certain small angles to each other which result in
phase interference in the most important parts of their diffraction
pattern, then one or more of these vehicles will produce a weak
signal. Close-packed vehicles lose part of their outline and the inner
vehicles show some loss of detection.
2. In none of the cases we have tested has the relative signal strength
fallen below 47 percent that of a single vehicle.
3. We feel that this test shows an interesting shortcoming of the process,
but represents a rather unusual condition that should give little
trouble in actual use.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A00230002001161
Section D-4
Multiple Object Filter
Discussion
In the normal use of the recognition process a spatial filter is made for
a single object and this filter is used to detect this object in the transparencies.
The readout then presents two kinds of information, the type of object and its
location in the picture area.
It is also possible to make a filter of several objects at the same time
called a multiple-object filter and recognize any of these. The readout is now
ambiguous, we can examine the entire transparency and obtain position information
without knowing which of the objects has been detected, or we can restrict the area
under examination to an area no greater than one object and obtain information on
which object was detected by the angle at which it reads out through the filter.
This is another kind of position information which when referred to the positions
of the objects in the filter tells which object was detected.
The use we have considered for multiple-object filters is a filter which
would contain several objects and thereby speed the search process. If the filter
contained ten objects and the use of the multiple-object filter did not change the
time constants of the search process, then a roll of film could be scanned for ten
objects in one-tenth the time it could be scanned with ten single-object filters
separately.
Purpose
The purpose of these tests was to test the features of a multiple-object
filter which would speed the process of scanning a large amount of film.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
1)-4-2
Previous Experience
In earlier experiments with character recognition it was shown that a filter
could be made for ten digits which would recognize any one of these with signal-to-
noise ratio of 40:1. Elsewhere, filters were made for 36 different characters which
produced satisfactory recognition at an unspecified signal-to-noise ratio.
It might therefore appear that ten vehicles could be placed in a multiple
object with the same results. Our experiments have shown that there are several
important differences between the two cases:
1. In the case of characters on an opaque background the contrast is
practically infinite and the "noise" is negligible. The multiple object
filter is always more complex and denser than a single object filter.
The strength of the recognition'signal obtained from a multiple-object
filter is therefore always less than the signal obtained with a single-
object filter. For characters on a noise-free background we can afford
loss of signal and still have an acceptable signal-to-noise ratio. For
aerial photographs which show a "noisy" background not much loss of
signal can be tolerated.
2. Characters are relatively simple shapes and produce simple spatial
filters. Vehicles and other aerial photograph "targets" tend to be more
complex and produce filters with a great deal of detail in them. For
example, the filter of a single tank appears to have as much or more
detail than a multiple-object filter for ten letters or numbers. This
means that multiple-object filters for vehicles will be packed with detail,
dense and show lower modulation efficiencies.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-4-3
3. In character recognition there is no orientation or scale scanning and
therefore no opportunity for "false-alarms" due to inverted letters or letters
of a different size. A filter made from several vehicles of different sizes,
all of which tend to be more or less rectangular or to have common features
such as turrets or guns will show partial recognition or "false alarms" as
the orientation is changed and the scale varied. In a sense, these many
false alarms tend to raise the noise level of the recognition image so that
in the case of aerial photographs, scanned with multiple-object filters the
signal is less and the noise greater so that the signal-to-noise goes down
very fast as objects are added to the filter.
Experiment with Ten Vehicles
Ten models of different army vehicles were selected and a spatial filter made of
them. The vehicles were of assorted sizes from a large tank to a jeep. The objects
were not arranged in straight TOWS or aligned parallel because it was felt that the
many parallel sides would act as a multiple line grating and produce a confusing effect
in the filter. Different filters were made which showed the optimum interference in
the 4th to the 10th diffraction order by attenuating the reference beam different amounts.
All of these filters tested very good and showed sharp reconstruction images. A
multiple-object filter tends to be denser than a single object filter and the high order
multiple-object filters were quite dense.
One of these multiple-object filters was tested in detail. When used to
recognize the object from which it was made it showed a very strong recognition spot
in the exact center of the pattern where no vehicle existed. This is interpreted as a
recognition of the entire pattern of several vehicles and since the spot is at the center
of the pattern it does not represent the recognition of any specific vehicle. The
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-4-4
intensity of the recognition was then measured for each individual vehicle by
masking off all the others in the object transparency. In this case, the recognition
spot also appeared at the center of the pattern. The results were as follows:
Signal Strength
(arbitrary units)
All vehicles 2500
German tank 2108 100
German Panzer tank-S 100
Rocket launcher 100
German tank T-34 110
10 wheel tractor unit 100
Tank M-47 60
Heavy gun carrier 45
German tank 077 20
Jeep 17
Tank M-48 90
It can be seen that the recognition .for the whole pattern was about three
times as high as the total recognition for all the objects in it. This is some
accumulative effect we cannot explain. Most of the vehicles were approximately the
same size and produced the same strength signal or 100 units. It is normal to expect
the jeep which is small to produce a lower signal. The area of the jeep image on the
transparency measures about one-fifth that of the tanks and so regarding the image
as a window, the light transmitted will be only one-fifth. Since the recognition
signal is derived from this light we would expect the signal to be about 20 units.
The low values for the tank M-47, German tank 077 and Heavy gun carrier are due to the
lighter color of the models. These had a different color ofucamouflage" paint which
photographed much darker on the negative transparency. The effect of dense images in
the transparency is like adding a neutral density filter to the object or its
recognition spot. The observed density of the images and strength of recognition
spots shows reasonable agreement.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-4-5
One feature that this test showed was that if multiple-object filters are
to be made, all of the objects should have approximately the same size and reflectance
or there will be a large disparity in the recognition signals for these objects.
An object transparency containing these same 10 vehicles, but in different
positions and on a "noisy" background was compared to the three best multiple-object
filters. The vehicles could be detected, but at such a low signal-to-noise ratio and
with so many false alarms that automatic recognition was considered impractical.
Typical values were:
Signal
Large vehicles 40
Small vehicles or dense images 16
Noise 10
False-alarms 20
The signal strength measured above is reported in the same units as in the
previous Table. The reason for the lower values in this case was the low-contrast
images in the simulated aerial photograph which produced weaker diffraction. Thus,
a high-contrast image of a tank might produce 100 units of signal, while a medium
contrast image produced 40 or less. The small and dense images produced still lower
signals for obvious reasons. The false-alarms came from all the partial recognitions
of similarly shaped vehicles at various combinations of orientation and scal,. Even
if there had been but one vehicle in the transparency we would probably have obtained
false-alarms for rectangular buildings, shadows of telephone poles and other normally
occurring objects which would have the same shape as tanks, turrets, guns or other parts
of the vehicles.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-4-6
Alignment Tolerances of a Multiple Object Filter
1. Displacement. The measured horizontal alignment tolerance was no
different than a single object filter. The vertical alignment tolerance was
slightly tighter, but within the usual experimental variation.
2. Rotation. The rotation tolerance for multiple object filter 1546
was half that of a good single object filter for the same object as shown below:
Rotation Tol. Single Rotation Tol. Mult.
Signal Object Filter Object Filter
100 .5 degree .3 degree
90 1.2 .7
80 1.9 1.0 II
50 3.9 2.0 II
This can be explained as due to the higher order of the filter. The
single object filter operated mostly in the 5th to 7th diffraction orders, the
multiple object filter operated mostly in the 10th to 16th diffraction orders.
The reason for this was that the multiple object filter showed very poor efficiency
and selectivity at a lower order.
Referring to the geometrical explanation for rotational alignment tolerance,
it can be seen that if the effective diffraction order is twice as high, these
orders will be twice as far from the axis and the same amount of rotation angle
will cause them to be misaligned twice as much, or to put it another way; the
angle error to produce the same signal will be half as much.
3. Magnification. The magnification tolerance for multiple object
filter 1546 was also about half that of a single object filter for the same object.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-4-7
Signal
Magnification Tol.
Single Object Filter
Magnification Tol.
Multiple Object Filter
100
1.0%
.2%
90
2.2%
.67
80
3.27
1.4%
50
6.07
2.67
The explanation again is probably the higher diffraction order which
apparently was required by the greater complexity of the filter.
Conclusion on Alignment Tolerance
1. The x and y displacement tolerance of multiple object filters is
the same as single object filters.
2. The rotation and magnificatign alignment tolerances are the same if
the filter is made in the same diffraction order.
3. In general, multiple object subjects will require a higher order filter.
Since the higher order filter has "spots" at greater radial distances from the axis
the rotation and magnification tolerances are reduced proportionately.
Effect on Time Constants
It is impossible to predict anything from the few experiments described.
However, if a ten object filter requires a diffraction order twice as high as a
single object filter, then the time to make the orientation and scale search will
each become twice as long and the product will be four times as long. Thus, the
time to make a complete search with a ten object filter will be four times as long
as the time to make the search with a single object filter. This is still consider-
ably less than the time required to make a search with ten single object filters.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-4-8
Conclusions for 10 Object Filter
1. Ten vehicles are too many for any multiple-object filter we have
been able to make because this filter produces a lower signal strength and
many false alarms.
2. The problem is different from the multiple-object character filter
because in that case we do not perform a scale and orientation search. For the
case described, a complete scan with the ten object filter would result in
about 60 false-alarms.
3. Due to changed alignment and time constants the ten object filter
would require four times as long to make a search as a single object filter.
Experiment with Two Vehicles
Two tank models M-47 and a German Panzer IV S were selected and a spatial
filter made for both of them. This filter was then used to recognize these
targets in a photograph containing these and eight other similar vehicles and
some model trees.
The two tanks in the filter recognized the similar tanks in the photograph
with a signal-to-noise level of ten to one which is considered satisfactory. False
alarms were also obtained for other vehicles as follows:
Signal Strength
M-47
100
Panzer IV S
90
M-48
40
M-48 inverted
30
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-4-9
Signal Strength
German Tank T-34
20
German Tank T-34 inverted
20
Heavy gun carrier
30
Heavy gun carrier inverted
20
Conclusions for Two-Object Filter
1. Two objects as complex as tanks can be handled by a multiple object
filter.
2. The alignment tolerance and time constants for search were the
same as for a single object filter.
3. In the experiment described, there could have been six false alarms.
Four of these were for very similar tanks, the other two were for an entirely
different vehicle that bears no obvious similarity to the tanks for which the
filter was made.
Experiment with Three Vehicles
The previous experiment was repeated, adding a German 2108 tank to the
spatial filter. The average signal-to-noise level for the detection of the tanks
was seven to one for the maximum noise. Twelve false alarms were obtained which
was twice the number obtained with the two-object filter.
The conclusion is that three or four tanks probably represent the maximum we
can put in a filter at the present time.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-4-10
General Conclusions on the Multiple-Object Filter
1. For simple objects such as characters on a noise-free background,
20 or 30 can be used, certainly ten are not too many. For more complex objects
such as vehicles and on a noisy background the present limit appears to be three
or four.
2. In these experiments the signal-to-noise ratio decreased as the number
of objects in the filter increased.
3. The number of false-alarms for similar objects increased as the number
of objects in the filter increased.
4. It is possible that improved techniques in making filters will
change this situation.
5. Multiple-object filters appear to required a higher order of
diffraction than single object filters and this will tighten the magnification
and orientation tolerances. This in turn increases the time required to make a
search with a multiple-object filter. The exact relation between number of objects
and increase in time is not yet known.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-5-1
Approved For Release 2005L05102 :DGIA-RDP78604770A002300020010-5
EFFECTS OF OBSCURATION - 1 Contrast
Purpose
Objects can be obscured or made difficult to find in a photograph if
they appear either in low-contrast with their surroundings or if the entire
photograph has a low-contrast.
Low-contrast aerial photographs can result when pictures are made on cloudy,
overcast or rainy days; or when the film is underexposed or improperly developed.
To test the effect of overall contrast we made a series of photographs identical
except for density range.
Experimental Program
Each photograph contained four tanks agains a "noisy" background. Included
in the photograph was a piece of white cardboard which acted as a brightness standard
to measure the contrast of the negative: The contrast ratio was defined as the
density difference between the image of the white standard and the unexposed edge
of the plate, and was therefore the total contrast range of the entire photograph,
and not the contrast of the tank against its background. Since the tanks and back-
ground were both fairly dark in the scene, the contrast of the tanks was considerably
less, about half the contrast of the photograph.
The low-contrast images were made by reducing the normal development of the
plates. It was necessary to mix special developers and use a two-bath system to
obtain satisfactory low-contrast images on the material we used.
As a normal photograph we used one with a total contrast of 1.5, since we
believe this is an average value for aerial photographs taken on a bright day. In
any case, this negative looked very much like aerial negatives we have seen.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-5-2
The 1.3 contrast negative was noticeably weaker, but still an acceptable
photograph.
The 1.1 density negative was the kind a photographer would call a "thin"
negative and the tanks were less obvious than in the higher contrast photographs.
The .7 contrast negative was quite thin and the images difficult to find.
The .3 contrast negative, unless looked at carefully appeared blank. The
tank images were very difficult to find. Most observers would find it difficult
to use such a negative.
The recognitioa signal for each tank and the noise background was measured
in each photograph.
Results
Contrast
Signal
Max. Noise
S/N
1.5
140 - 160
19
7:1
1.3
100 - 120
15
7:1
1.1
58- 80
8
7:1
.7
27 - 40
4
7:1
.3
17 - 25
3
7:1
These figures show that the inherent noise in the film decreases in the same
ratio as the contrast, so that low-contrast images show practically the same signal-
to-noise as high contrast images. These are exceptions to this of course. All of
these photographs had practically perfect surfaces with no scratches or defects.
The diffraction caused by a scratch is constant and therefore a scratch is much
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-5-3
more serious on a low-contrast negative than a normal contrast negative. In one
test a fingerprint was put on the back of the glass plate. The noise level rose
from 4 units to 27 units which was equal to the strength of the weakest signal in
this case. Therefore, while it is feasible to find images in very low-contrast
photographs, it may prove to be impractical if the film is scratched or finger-
marked.
These tests indicate that a machine should be able to find low-contrast images
as easily as high-contrast images, but these tests were performed manually and it
was necessary to scan the field for recognition images by eye and then align the
measuring microscope with them. The actual tests were much more difficult to perform
on the low-contrast negatives because of the difficulty of locating and orienting
the low-contrast images by eye. Figure 1 shows a low-contrast photograph.
Conclusions
1. If the negatives are in perfect condition, with no scratches or flaws, the
automatic recognition process will locate images of objects at extremely low-contrast
and at the same signal-to-noise ratio as obtained in normal contrast negatives. This
potentially makes the process suitable for locating images too weak to see. In this
experiment good recognition was obtained for images that had a density difference
from this background of .15 or less.
2. The intensity of the recognition signal is less in low-contrast photographs,
therefore defects which raise the noise level are relatively more serious.
3. The results of this experiment cannot be directly related to a normal
contrast photograph in which the targets appear in low-contrast due to camouflage
or other reasons. In this case the object is at a low-contrast and will produce a
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Overall Contrast Contrast .15
Figure 1 - Low Contrast
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-5-5
weak signal, but the average noise level will be that of a high-contrast photograph.
For example, if the target were camouflaged to show a density difference of only .15
from its surroundings, but the photograph as a whole showed normal contrast, then
we would expect the signal and maximum noise to be about equal.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 9/Q .9A-RDP781304770A0023000201)16=t
EFFECTS OF OBSCURATION - 2 Camouflage
Purpose
This test was to determine the effect of a confusing background, especially
one in which the major color was the same as the vehicles and therefore provided
little or no contrast, while the background itself contained many similar patterns
in high contrast. This was the principle of "dazzle camouflage" used in World War I.
Discussion
The tanks used in this test were a dusty olive drab color. Four of them were
placed on a piece of "camouflage cloth" purchased from a surplus store. This material
has an irregular design in three colors, one of which was an olive drab that matched
the tanks almost perfectly. Also, the general pattern of the cloth produced differently
colored areas each about the size of a model tank. The photograph was diffusely
illuminated so there would be no shadows to outline the tanks. The resulting photograph
was a good example of "dazzle camouflage" in that the tanks were difficult to find and
the eye was misled by the overall pattern.
Results
When the area around any tank was masked the intensity of the recognition signal
for that tank was normal for a low-contrast target. Diffusely illuminated objects
lack the sharp shadows which are one of the chief causes of diffraction and an
important aid in recognition. It the normal recognition intensity for a sunlighted
tank was 100 units, the value for a diffusely lighted tank simulating a photograph
taken on a cloudy day would be 30 units. The four different tanks gave recognition
signals between 23 and 40 units. However, when the area around the tank was uncovered
and the entire photograph used, the intensity of the recognition spot dropped to as
low as 8 units. This is the first time that this has been observed. Normally, if
the area around an object is masked, the intensity of the recognition spot does not
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-5-7
change enough to measure. In this case the average size of the camouflage color
areas was close to the size of the tanks and we theorize that the diffraction of these
many areas produced a strong "noise" diffraction in spatial frequencies close to that
of the object. Apparently the energy in these frequencies optically interferred with
the energy diffracted by the tanks and reduced their intensity. The same result was
noticed for all four tank images.
The noise in this photograph was very strong and "spotty" so that it was
difficult to choose the maximum noise to measure, but values as high as 23 units were
found. The result of viewing the photograph as a whole was as follows:
Signals for tanks 6 - 8 units
Maximum noise 23 units
Average noise 8 units
In other words, the signals for the tanks in this case were below the average
noise level and therefore undetectable except when the operator knew where to look
for them. Figure 2 shows the photograph used.
Conclusions
1. Dazzle camouflage produces strong diffraction noise. If the average
frequency of this noise is close to that of the chief spatial frequency of the target,
then the target may be obscured by noise and be undetectable by this process.
2. For reasons not entirely understood, the presence of a confusing background
actually reduced the intensity of the recognition signals.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Appr ved EQr.,ReleaseP78 002300020010-5
Camouflage - Dazzle Effect (Tank at upper right was isolated as shown after
Approved For Release 2005/0g/s0t2s: Clik--PRBIW8i304770A002300020010-5
Figure 2
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-5-9
3. The target in these photographs was diffusely lighted (no shadows) and
showed a maximum density variation of .55 which simulated the situation of a cloudy
day.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020a05510
Section D-5-3
EFFECTS OF OBSCURATION - 3 Low Contrast
Purpose
Previous tests showed effects of a low-contrast image on a plain "non-noisy"
background and the effects of a low-contrast image on a background of noise having
a high intensity and a mean spectral density close to that of the signal. This
test was to find the effect of a low-contrast target on a background of high
frequency noise of the same intensity.
Discussion
Four similar tank models were photographed on a finely mottled background.
The average contrast of the tank images and background were both close to density
.65 and the scene was flat lighted to eliminate shadows. As result, the entire
photograph is in low-contrast and the tanks are fairly difficult to see.
Results
The recognition signals for the four tank images varied between 70 and 100
units. The background showed a strong noise area which had a maximum value as
high as 33 units though most of the noise measured about 10 units. The cause of
the high noise area was a scratch in the emulsion. This shows the severe effect
that scratches can have, especially in low-contrast pictures.
Results with Scratch
Signal
Noise
S/N
70
- 100
33
2.1:1
Results without Scratch
70
- 100
10
7:1
Conclusions
1. All low-contrast subjects give a lower signal-to-noise ratio than high
contrast subjects. The addition of defects to the film, a scratch in this case is
relatively more serious in the case of low-contrast subjects.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-5-11
2. In this case where the "noise" background had the same contrast as the
target, the detection was fair, much better than the case where the noise was at
high contrast.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020019:?_ 12
Section D-5-4
EFFECTS OF OBSCURATION - 4 Noise Background
Purpose
This test was to determine the effect of a high-contrast noisy background
on the recognition of normal contrast objects.
Discussion
Four similar tanks were photographed on a spotted background made by putting
dark colored random size spots on brown paper. The result was what might be called
a "spattered paint" effect. In the photograph the tanks appear at the contrast they
would have in a normal aerial photograph and they are seen the very noisy background
or random size spots. All of the spots are smaller than a tank and therefore the
noise spectrum consists chiefly of frequencies higher than the tank outline, but
about the same as the tank details.
Results
The intensity of the signal for these normal contrast tank images was about
five times the signal obtained for low-contrast images in the previous tests.
While the background produced strong "noise" at random frequencies it did not
interfere with the detection of the signals.
Signal Noise S/N
350 - 520 15 22:1
Conclusions
1. If the targets appear in average contrast, then they can be detected
against a "noisy" or confusing background, especially if the peak noise spatial
frequency is not the same as that of the targets.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78B04770A00230002001191_13
Section D-5-5
EFFECTS OF OBSCURATION - 5 9verlap
Purpose
This test was to determine the detectability of objects when they are
partly concealed by opaque covering.
Theoretical_pasLground
It is clear from theory that the strength of the recognition signal should
not decrease linearly with the unobscured area when an object is partly covered.
The exact relation is unknown at present and may have to be determined experimentally.
As long as five years ago the mathematician Van Heerden tried to calculate two cases
for us and reported that the mathematics was not simple nor the results definite;
but to his best guess, when 90 percent of the object was visible the signal should
be practically the same os for an unobscured object and when 50 percent of the object
was visible the signal should be close to .9 the normal signal. Even when only 10
percent of the object was visible the signal should be fairly high. These calculations
were made on the basis of letters and words of the English language and do not
necessarily apply to other objects, though the general conclusions probably do.
If this proves to be true, then this system would be useful for detecting
partly hidden objects and should find them even when only a small portion is visible,
or in more practical terms until the reduced signal approaches the noise level. As
in all these other situations, the ability to detect hidden objects will depend on
the noise background.
Discussion
Four similar model tanks were photographed on a "noisy" background of rough
sand. One tank was not concealed and used as a control; the other three were covered
by model trees made of lichen and a different part of each tank was covered. There
were four photA
ISPfe1edn12161teRelebi4120?0e5/0E2b96tA4DP78130477OM70710001210010-5 covered,
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-5-14
in the others, 40, 60 and 80 percent. Figure 3 shows a typical photograph.
Results
1.
Tanks 20 percent covered
Signal
a. Control (not covered)
100
b.
Left rear covered
100
c.
Left front covered
90
d.
Rear end covered
90
2.
Tanks 40 percent covered
a.
Control
100
b.
Left side covered
80
c.
Front end covered
83
d.
Rear end covered
93
3.
Tanks 60 percent covered
a.
Control
100
b.
Left side covered
97
c.
Front and left side covered
70
d.
Rear end covered
90
4.
Tanks 80 percent covered
a.
Control
100
b.
Only little of left side visible
58
c.
Only little of right side visible
50
d.
Only left front visible
60
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Obscuration - 80 percent Concealed
Approved For Release 2005/05/0?,qA-FDP781304770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
D-5-16
These photographs are not strictly comparable because the noise level was
not constant. For reasons not known entirely at present, the noise level in all
of them was higher than normal and increased from about 10 units in photograph
1 to 15 units in photograph 3 and 20 units in photograph 4. This may be partly due
to the fact that it was necessary to move the lights in the last picture and this
may have put more shadow detail or noise on the sand background. If this is taken
in account then the average signal-to-noise ratio for the four cases is:
Percent Concealed
Signal/Noise
20
10 :1
40
8.5:1
60
6 :1
80
2.8:1
There should have been no difference in the noise background of the four
photographs and the only variable should have been the degree of concealment. Also,
the noise level of all the pictures is about twice as high as that of most of our
photographs, so that if a normal noise level had been obtained the signal-to-noise
values would be at least twice that shown.
Conclusions
1. The recognition signal is not proportional to the visible area of the
object. It is not logical to expect half of the object to give a half strength signal.
2. The present tests were confused somewhat by a variable noise background, but
they indicate that good recognition should result when 60 percent or more of the target is
Some recognition was obtained for targets 80 percent obscured.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5 n- 5 - l 7
EFFECTS OF OBSCURATION - General Conclusions
I. The recognition signal comes from light diffracted by the image of the
target and then again selectively diffracted by the spatial filter. In all
aerial photographs it is necessary to detect this signal against a background
noise which results from all the random shapes and textures which give rise T:o
diffracted light representing many spatial frequencies in the scene. In gener ,
a natural background does not seem to present a high noise level at frequencies
that interfere with detection of man-made objects. The introduction of a baekgr:u d
filled with human artifacts such as roads and buildings produces a higher level 1
noise at frequencies close to those of typical man-made objects that might be
considered as targets. Therefore, the effective signal-to-noise obtainable will
probably be less in urban scenes than rural scenes.
2. Since the recognition signal is basically derived from light diffracted by
the image of the target, any circumstance which reduces this diffraction will resm. t
in loss of signal. Typical circumstances are low-contrast, lack of sharpness anc
obscuration.
a. Overall low-contrast of the entire photograph is not a problem
because in this condition the noise diffraction reduces in the
same ratio as the target diffraction.
b. A low-contrast object on a normal-contrast background is
difficult to detect.
c. A low-contrast target on a high-contrast background is impossible
to detect when the spatial frequencies of the background or
camouflage are close to that of the target.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Section D-6
Effect of Target Aspect
Introduction
Objects on the ground which appear at the center of a vertical photograph
are seen exactly from above. The same object at the edge of a wide-angle phot-
graph will will show a different aspect or perspective and portions of the sides or
ends will appear in the photograph. Oblique and panoramic photographs show
objects from many aspects.
If we assume that the spatial filter for a target is made from an image
showing a vertical aspect, then it can be predicted that objects appearing in
other aspects will not match the filter perfectly and there will be some loss
of recognition.
The purpose of this test was to determine the effect of reasonable aspect
angles on recognition.
Experiment
Four model tanks, turned in different directions were photographed with a
long focal length lens. One picture was made directly above and others at
various angles. Each of the tanks in any one photograph was seen from the same
nspect angle, though due to its orientation each tank presented a different
appearance. The signal-to-noise ratio was measured for each photograph and
also the relative signal strength of each tank in each photograph.
In the first test, normal recognition signals were obtained for the vertical
images, fair results for ten degrees of obliquity but poor and erratic results ter
Larger angles.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A0023000200107_,
Section D-7
Effect of Shadows
Introduction
On a cloudy or rainy day there are no shadows in aerial photographs. On
sunny days there are shadows on the ground and for any particular object the
size of the shadow will vary with the position of the sun. For high sun angles
the shadows are small and close to the objects and serve to outline them and make
them more visible. It appears that these small, sharp shadows improve the
recognition signal because we have always obtained better results with such
photographs than ones taken with diffuse light to simulate a cloudy day.
However, as the sun angle becomes lower the shadows become larger and
extend to greater distances and often assume peculiar shapes. Since the shape
of a shadow depends on the sun angle and orientation of the target, the shape
is quite variable. It appears that these shadows do not aid the recognition
process and may hinder it to some degree.
A second effect of a low sun angle is loss of detail on the target. When
lighted from above, all of the detail on the top surface is visible and hence
comparable to the filter which is assumed to be made from a diffusely illuminated
object. As the sun angle becomes low many of the areas on the top surface fall
into shadows and the detail on them disappears.
Experiment
Four model tanks turned in different directions were illuminated with a
lamp which simulated the shadow contrast produced by sunlight. Vertical photo-
graphs were made with the "sun" at various angles to the camera. In any one
photograph the four tanks were all illuminated at the same angle, but showed
different shadows due to their orientation.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
Section D
Conclusions
Only a few of the possible image variables were studied and these were
measured by means of laboratory made simulated aerial photographs.
The general conclusion is that the recognition process should be capable
of finding any object a human operator would normally find. It is also possible
to have objects too small to detect with a particular instrument, and certain
types of camouflage confuse the recognition process. On the other hand, the
process has shown good performance on low-contrast and partially hidden objects.
An exception is deliberate camouflage which may alert a human operator to look
for additional clues. Human beings have intelligence to cope with all sorts of
difficult situations when necessary, but the recognition process has no i-itell-
igence.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Section E-1
Test of Recognition on Aerial Film - I
Introduction
Aerial camera film was tested for recognition of vehicles and airplanes
appearing in the pictures. Very poor recognition was obtain on the first roll
of film in which the intensity of the recognition spots was approximately equal
to the strongest noise in the scene. This is considered unsatisfactory for
automatic recognition.
A preliminary analysis of the cause was made and the experiments are
described.
The film consisted of a strip of 78 9 x 9 inch frames and 33 panoramic
negatives.
All of the film was apparently a reversal processed duplicate negative.
While these are very good duplicates, it was our idea that the recognition process
would work best on the original negative.
As will be described later in the experiments, we were unable to obtain
good recognition in any of the frames tested. Definite recognition was obtained
in almost every case, but the recognition spots were never as much as 1.5 times
the maximum noise in the picture. Since we believe that a signal-to-noise at
least 2:1 and preferably 5:1 or higher is necessary for automatic operation, the
recognition was considered unsatisfactory.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
- 1-3
4. Use of a less than perfect optical system.
5. Use of less than perfect liquid gates.
Each of these will be explained in some detail.
1. Loss of low-contrast diffraction detail. The earliest spatial filtering
experiments conducted by Marechal and Toraldo di Franca showed that it was possible
for a negative to contain an image of the light diffracted by objects too small to
be imaged by a lens. This diffraction detail was recorded over a large area at
very low contrast something like a hologram. The spatial filtering apparatus
duplicated the Abbe condition where the geometrical image and the reconstructed
diffraction image were superimposed and it was then possible to see fine detail
in the processed picture that could not be found in the original negative. We
have assumed that something of this sort took place when original negatives were
used as the objects for recognition experiments. We have also assumed that this
detail was recorded at such a low contrast that it would be practically impossible
to copy it.
2. Loss of sharpness. There is inevitable loss of sharpness in any
copying process. The best diffraction is produced by sharp edges.
3. Weak diffraction due to grain noise. Edges that are diffuse or rough
produce weak and complex diffraction compared to sharp, high contrast edges. The
copying process inevitably superimposes the grain pattern of the copy film on that
the original film resulting in a grain pattern that is worse than either film above.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
E- 1-5
Description of Experiments
Several frames were selected for recognition experiments. Using two frames
which showed the same or a similar vehicle or airplane, a spatial filter was maCe
from the object in one frame and used to recognize the same object on another. in
no case was a filter used to recognize the object from which it was made. We fee,
that to do this would be dishonest.
The spatial filters were made in several ways:
1. From objects taken from adjacent frames as described above.
2. From models of similar vehicles.
3. From drawings made from the objects in the aerial films.
The filters were aligned with the axis and a search made for the objects.
In every case a recognition spot was obtained, but it was so weak in comparison to
the general noise level that it could only be recognized because the operator knew wht-re
to look and what to look for. Recognition spots have a distinctive appearance and
surround. For example, a very weak recognition spot for a rectangular tank shows a
small circular spot with four thin rays aligned with the principle dimensions,a dark
area representing the area of the tank and a faint bright outline of the tank itself.
These visual clues aid a human operator to find the recognition spot though they do
not aid in machine recognition.
Experiment 1. Images of a Patton tank were compared to a filter made from
another frame. Using either the full aperture, 2 1/4 inch aperture or 1 inch apertur-
of the film with or without a liquid gate the recognition spot measured about.8 to 1.
times as high as the highest noise in the picture. This was about 30 times as high a
the average noise.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02: CIA-RDP78604770A002300020010-5
E-1-7
Experiment 9. Images of the Beechcraft airplane were compared to a filter
made from a photograph of a model. The recognition of high contrast images was
fair, 2-3 times above the noise level, but low contrast images gave recognition
at the noise level. High contrast images were produced when the aircraft,
normally painted white, was on a dark runway. In this case the outline was very
clear and there were no confusing shadows. Low contrast images were produced when
the aircraft was on a clean concrete runway in which case all that showed was the
shadow which was a different shape than the airplane; or when the aircraft was
painted a dark color and standing on a dark runway.
False alarms equal in intensity to about half the recognition signal were
produced by other smaller airplanes, even single engine aircraft.
Experiment 10. The previous experiment was repeated using spatial filters
showing a higher order of diffraction. The filters used in Experiment 9 were 1st
and 2nd order. In this experiment several filters combining 3rd and 4th and 5th
and 6th orders were tried. The general noise level and the false alarms were
reduced but not enough to insure reliable detection. In this experiment it
appeared that there was a direct relation between the contrast visibility of the
airplanes and the strength of the recognition signal. As we understand it, this
is not supposed to be true in a properly operating system, though there will be
some reduction of recognition with loss of contrast.
Experiment 11. Images of the Beechcraft airplane was compared to a filter
made from an enlarged section an aerial film in which the airplane was cut out
with scissors. This is equivalent to a drawing of the object. Actually, two
cutouts were made of different images. The results were practically the same as
the experiment using the model as an object for the spatial filter.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
F- 1-9
In these experiments, vehicles had to be isolated from the aerial film.
Taking as an example a tank which appeared dark on a light background the negative
showed a transparent tank on a darker transparent background. This image was contact
printed on a glass plate to make a positive. The glass plate was fine-grained and
developed to give the same contrast as the negative. This positive was then contact
printed to make a negative and this image was the one used to make the filter. On
large images such as 5 mm tanks, the background around the vehicle was painted with
opaque paint and masked with tape to produce a plate in which the only transparent im
was the vehicle. This plate was then used as the object to make the spatial filter.
When these images were enlarged 6 times they looked good, but when enlarged 40 times
the outlines were indistinct and the image grainy. It could be seen that the maior
detail in the image was shadow detail, and of course,this varied with the aspect of the
sun. Therefore when filters made from these vehicles were used to recognize other
vehicles, the shadow detail was quite different.
On the other hand, our laboratory pictures could be enlarged 40 times and
still remain sharp and grainless,and due to the fact that they were diffusely lighted_
the shadow detail was not distinctive of any particular angle of lighting.
In the case of models it is necessary to eliminate any shiny surfaces which
would produce highlights distinctive of the lighting. We therefore either sandblasted
the models or spray painted them with dull gray paint. If it is necessary to produce
a very opaque background in the photograph it is not enough to make the picture on
a white background, in general it will be found that sufficient exposure to produce a
background of density 3.0 plus will result in a fairly high density vehicle image
also. Such poorly transmitting object transparencies are difficult subjects for
spatial filters and require long exposures. Instead, the model is placed on trans-
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
E-.-11
Conclusions First Roll Aerial Film
1. Recognition obtained on the submitted aerial film was poor. This
was chiefly due to loss of signal strength and not gain in noise, though in at
least two instances minor damage to the film resulted in high local noise.
2. The film was apparently reversal processed duplicate negative and showed
no detectable relief image.
3. The diffraction images produced by the objects on the film were very
weak compared to images made in our laboratory.
4. The spatial filters do not appear to be at fault.
5. Recognition on contact prints from the aerial negatives was considerably
improved, but still less than anticipated.
6. Spatial filters made from sharp originals such as models and cut-
outs gave slightly better recognition than filters made from images taken from
the aerial film.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/c0t2inCIDP781304770A00230002001015 2-1
SECOND ROLL OF AERIAL FILM
Purpose
These tests were to determine the detectability of objects on aerial film.
Introduction
Earlier tests made on duplicate negative film showed poor results. The flim
was fine-grained, the images were reasonably sharp and in good contrast, but th.,
images on the film showed very weak diffraction. We therefore requested an ori4in,,.
negative. The new film used in the tests reported in this section was 5 inche wo,
and designated "High Resolution Test Film, Can 24".
The film was very long, containing many duplicate and overlapping frames
produced by the camera plane flying over the same course several times. Two framo
near the end, Numbers 47366 and 47365 were chosen for tests. These two overlannins
pictures showed a corner of an airfield with several airplanes of three types. -ffi(
images were sharp, but somewhat smaller than desirable. As shown in earlier tet:.
the apparatus we use gives optimum recognition for objects between the sizes of lo
and 2.5 mm, or for larger objects if they contain small details. The sizes of he
three types of airplanes on the negative were:
Airplane Wingspan
Small ,97 mm
Medium 1.90 mm
Large 2.12 mm
We would therefore predict on the basis of previous tests that the small airbInne!.
would give weak or marginal signals, while the two larger planes would provide a
reasonably satisfactory signal. For best results, this film should have been
examined in an instrument with shorter focal length lenses, or the negative could
have been enlarged to fit the apparatus. neither was done.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A00230002001015 -3
have been used to search the original negative which appears to be the optimum
condition for two-beam spatial filters. However, to do this would have required
a vignetting mask around the object. When the airplane is dark and the surrounding
transparent there has to be an edge or border to the illuminated transparency. It
this edge is sharp, square for example, it will produce the diffraction of a sqinr
as well as the diffraction of the airplane. The resulting filter will then attempt
to recognize an airplane in a square. To avoid this and to prevent diffraction of
the edges of the field it is necessary to use a vignetting mask, a glass plate
film with a clear spot at the center and a gradually tapering density out to thc e,
where it becomes totally opaque or at least density 3.0. No such mask was avallabl,
at the start of the tests and so in the first experiments the spatial filter wn= ma,
from a positive and used to recognize airplanes on a negative film.
This situation, which frequently arises is known as using a filter of the
oppositive polarity. In previous experiments with characters, the polarity of rio
filter appeared to make little difference. We have assumed that for more compicx
objects, the best recognition would be obtained if the film and filter had the
same polarity. The situation is complicated, the amplitude portion of the filter
certainly shows polarity, but at present we believe the phase portion does not;
there is no difference between negative and positive phase. Therefore, when we
speak of a filter of the opposite polarity, probably only one part of it is opposir.
Making the Vignetting Filter
As explained above, this item is required to make spatial filters of objects
that are dark and have transparent surroundings. For a 2 mm diameter airplane
image the filter should have a transparent center 2 mm in diameter and gradually
decreasing transmission to practically zero at a diameter of about 10 mm. Making
this filter proved to be difficult. Three different means were used and severai
tries of each one and none of them produced a really good filter. The best one was
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010-5
Most of the airplanes were on a high contrast background and were easy to
see (though quite small), one airplane was on a white concrete runway apparently
in motion and perhaps airbourne. It showed very low contrast against the runway
and a different aspect (foreshortened). Poor recognition was obtained, the
strength of the signal was only one-fifth that of the other airplanes.
The small airplanes gave weak recognition signals due to their sma1 size.
The signals were about twice the maximum noise.
When tested with three different "negative" spatial filters made wath a
rather poor vignetting mask, the large and medium airplanes gave good recogniti?
The ratio of the signal to the maximum noise was 6 to 1 and the ratio ot signal ;.c
most of the noise was 100 to 1. False alarms similar to the earlier tests were
also obtained. There seemed to be very little difference between the performance
of the two sets of filters.
Figure la shows the area covered by the negative and the actual size of
the airplane images. Figure lb is an enlargement of a small area to show the
quality of the airplane images.
Figure 2 is an enlargement of the hangar area showing the relative positions
of the four large airplanes.
Figure 3 is an enlargement of the recognition image of this same area and
shows four recognition images where the airplanes were located. This photograph
was made by stopping the scale and orientation search at three different times
and making successive exposures on the same film.
Liquid Gates
Both surfaces of the aerial negative 'were crisscrossed with scratches, some
as much as half a micron deep. There were also a number of other minor surface
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Figure 1 a Contact Print of Aerial Negative
Figure 1 b Portion of Above Enlarged 16 Times to Show Image Quality
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Figure 1
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
E-2-7
Enlargement of Hangar Area
Approved For Release 2005/05/012i:ggA-FpP781304770A002300020010-5
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
Recognition Images for Four Planes - Area is Same as Figure 2
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5
tirf, 3
Approved For Release 2005/05/02 : CIA-RDP781304770A002300020010142-9
defects. Supposedly, these defects produce diffraction which contributes to the
noise, and one of the purposes of a liquid gate is to eliminate the diffraction
effects of these scratches. The experiment was repeated with the negative in a
gate filled with mineral oil and there was no detectable difference either in the
absolute signal strength or the noise. This is the first time this has been
noticed. Usually, immersion in a liquid gate reduces the signal by suppressing
the relief image and also reduces the noise.
Relief in the Negative
Negative 47366 was examined for relief. It was normal for a fine-grained
image. The edges of the runways showed a relief of .15 microns, the airplanes
showed about the same. Some high-contrast details showed .30 micron. In contrast
to this the back side of the negative showed many scratches at least .5 micron deep.
Conclusions
1. Large (2 mm) and medium (1.9 mm) airplanes were easily detected at a
satisfactory signal-to-noise ratio as long as they were on contrasting
backgrounds.
2. A similar plane on a very low-contrast background would probably not
have been detected.
3. Small (.9 mm) airplanes were detected with a marginal signal-to-noise
ratio. These objects are really too small for our equipment.
4. The use of a liquid gate in this case made no difference.
5. The polarity of the filter seemed to make little difference. This
means that negative images on the film were detected almost as well
with filters made from negatives as positives. Actually, better
results were obtained with positives and the probable reason is our
limited experience in working with the vignetting filter.
Approved For Release 2005/05/02 : CIA-RDP78604770A002300020010-5