CAMOUFLAGE DESIGN

8/11/2019 CAMOUFLAGE DESIGN

1/80

TF-- HNIGAL REPOR r/ ....

MODIFCATIONS TO IMPROVE--£ 0DATA ACQUISITION

FQRAND ANALYSISSFOR CAMOUFLAGE DESIGN

NANCY MONASIU

DcILOG, INC.

I•fE hR RELEcSTF-f UMtOADETUNLIMWITD,~83 NMELVUJJJ, N.Y. 1170703 JMNUMWY 006A

UNITEDSTATESARMYNATIC;KREMSEARCH& DEVELOPMENTLABORATORIES

W~ICK M4ASSACHUSETTS0176000*R0VED FOR $OU8LCRELEASE DISTR1PUTICNI UNLUMOTD.

0020' ()o.


2/80

Approvedfor public nesseej Aistribution m 'litat.1

citation of trteo nern In this report oes wtoycstitute a&off i i a Indorsemmt or Rqw' ml of the

Detry Ureot tAwn no loiger lbflftU D o o tnettgn It to the originator.

11 1-limIM O M MI


3/80

UnclassifiedSECURITY CLASSIFICATION OF THIS PAGE (Wmon Dais 6n.'016P040______________

REPORTDOCUMENTATIONPAGE BEFORE COMPLETING FORMT.REPORT NUMMIN .GOVT ACCESSION NO. S. RECIPIENT S CATALOG NUMBERt

NATICK/ TR- 83/ Oll_____. ___________

MODIFCATIOS 0 TOIPOE DATA ACQUISI.TICN' ANDTyoRE:TPRoCVROANAYSIO CAOUFAGEDESIGN F n l R p r

6,PERFORMINGORG. REPORT NUMBER_________________________________________ DecilogReport_#245

7. AU THORt(o) I, CONTRACT OR GRANT NUWMER(a)

Arthur A. Gold# J. Richard Goldgrabeng C. Thomas DAAK6O-79-C-0072Goldsmithp Nancy Monastero (000005)

I PERFORMING ORG0ANIZATION NAME AND ADDRESS 10. PROGR~AM 9LEM9 , R O E T TASK

9 ~DECILOI, NC. .6PlCtr UMER555 Eroadhollow Road 16 2 H 8 ~ lMelville, NY 11747 A r

It. CONTROLLING OFFICE NAME AND ADDRESS 12, 04000PRT CATSUS Army Natick R&D Laboratories JAN 1983Attng DRDNA-ITC 7j7 NAUMBER oF PAGESNatick# MA 01760 _______________77

14. MONITORING AGSENY -NAMiTA ADORESS1(it dilf Ieit Iflat Ccullrailind Offlice) IS. SECURITY CL AlS. (of this repart)

UNCLASSIFIEDFil la . , %9CkOASSFICATIONIDOWNORADING

1S, DISTRIBUTION STATEtMCNT (of this Report)

Approved for public release; distribution unlimited

I7, DISTRIBUTION STATEMENT (of the abstract eiMtefedn 911661 20, Il dil OP.RI A o 11e11116)

Approved for public release; distribution unlimited

IS, SUPPLEMENTARY NOTES

I#- KEY WORDS (C.,tINNUOOAt

eV*?ee eidsitnecessary

and idenltify by black number)CAMOUFLAGE ELECTRO.OPTIC DETECTION SYSTEMSTERRAIN S VEGETATIONPATTERNS VISIBLE SPECTRAFIELD CLOTHING COLORSHELTERS SPECTRAL '

20. Ao~rIACr CWAAWems e o e c h i l ls M s e v s a imdtaltdeb, blocknwein) The'.report re-scri es theresults of modifications to a software program which clusters digitized, multi.spectral photographs of natural vegetative terrains into facsimiles of theoriginal scenes in 3, 4# or 5 colors in CIELAB notation. Tasks that were

addressed included optimization of the clustering to remove any bias introducedby the highly efficient histogram preprocessing scheme; evaluation of dataseparability parameters and a determination of the minimum numer of domainsnecessary to represent a scene; the a b l t 9 o a n l z n mont of the origi-scene; and the ability to plot each resultin domainal dvuly

I*"', 1473 noTor i Nov soIS OMO40LETIE UnclassifiedSECURITY CLASSIFICATION OF KtHI PAGE When Dotes Sit wed


4/80

Summary

This report presents the results of a modification to a research programto acquire data on the spectral and spatial character is t ics of natural vegetative

terrains and to develop methodologies for the analysis of these data as an aidIn the design of camouflage patterns for f ie ld clothing and large cloth shelters.

The program has been concerned with camouflage for the visible spectrum and for

human observers.

Technical aspects In the development of modifications of the existing soft-ware for data analysis are summarized and recommendations for further software

development are presented.

The following tasks were addressed:

1. Optimization of the clustering in the CIE 1976 (L*a*b*) color space toremove any blas Introduced by the highly efficient.histogram preprocessing scheme.

2. Evaluation of data separability parameters and a determination of the

minimum number of color domains necessary to represent a scene.

3. Modiflcatlon of the terrain analysis joftware so that the user can

specify any rectangular segment of the original scene as the area of Interest.

i4. Modifications of the symbol plotting software to enable individual

domain plotting.

All software has been Implemented on the UNIVAC 1106 computer located at the

Natick Research and Development Laboratories.

o...4.

1 "


5/80

PREFACE

The reported work was performed for US Army Natick R&D Laboratories under

Contract No. DAAK60-79-C-0072 (Modification No. P00005) with Mr. Alvin 0.

Ramsley, Project Officer. The Decilog effort was ably led by J. Richard

Goldgraban. This work is part of Project IL62723AH98, Clothing, Equipment$

and Shelter Technology; Task AD, Passive Countersurveillance Measures for theIndividual Soldier.

2


6/80

TABLE OF CONTENTS

PAGESUMMARY, .. . . . . . . . . ..v # 9 . . . ..

PREFACE. . ,...

LIST OF FIGURES. . . . . . . . . . . . . . . . . . . . 4

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . 6I. OVERVIEW . . . . . . . . 7

a. Background . . . . . . . . . . . . . . . . . 7b. Task Requirements for This Phase . . . . . . . ..

(1) Optimization of Clustering. ... . . . . . . . . . . . . 8(2) Data Separability . . . . . . . . . . . . . . . . . . . 9(3) Scene Segmentation. e s s . . . . , . t.... 9

(4) Domain Plotting . . . . . . . . . . . . . . . . . . . . . 9

2. OPTIMIZATION OF CLUSTERING , . . . . . . . . .... . . . . . 10a. General Discussion . 0. . , . . . . . . . . . . . . . 0b. Valldatlon of Subroutine OPTIM . . . . 1....... , 4

3. DATA SEPARABILITY . . . . . a . a . . . . . a . . . . . a 17a. Number of Damains Required 1. ..... . .... . . . . 7

b. Goodness of Domain%. 384. SCENE SEGMENTATION . . . . . . .. . . . 42

a. General Discussion . .... . ....... 42b. Symbol Plots of Scene Segments . . . . . . . . ....... 43

5. DOMAIN PLOTTING . .. .. . . . . . . . a a . 50

a. Individual Domain Symbol Plots . . .. a.. ... ... 50

b. Contour Plotting . . . . . . . . . . . . . . . a . .a 506. COMMENTS ANDRECOMMENDATIONS .a. . . a . . . . . . .a. . . . 65

a. Scene Digitization . a a a . . . a. .. . . . a. .. . 65b. Interactive Data Analysis. . . . . . . . . . . .. 65c. Alternative Data Collection a. . . . ... . 66

REFERENCES . . . . . . . . . a a . . . . . . .a. 67

Appendix. Outputs of Terrain Analysis Software Showing Stat is t ical

Data ........................ 68

3.


7/80

LIST OF FIGURES

PAGE

Figure I Subroutine OPTIM . . . . . . . . . . . . . . . . . . . 11Figure 2 Subroutine REDOM. . . . . . . . . . . . . . . . . . 12

Figure 3 Centrold Migration During Iterations of OPTIM ..... 16Figure 4 Four Idealized Domains ........ . ........ 18

Figure 5 Scene 10 (Beta vs. Number of Domains) . . . . ..... 22

Figure 6 Scene 3 (Beta vs. Number of Domains). . . . . . . . . . 23

Figure 7 Scene 4 (Beta vs. Number of Domains). . . . . . . . . 24

Figure 8 Scene 11 (Beta vs. Number of Domains) . . . . . . . . . 25Figure 9 Scene 12 (Beta vs. Number of Domains) .... . .. . 26Figure 10 Scene 13 (Beta vs. Number of Domains) . . . . . . 27

Figure I1 Scene 14 (Beta vs. Number of Domains) . . . ...... 28

Figure 12 Scene 21 (Beta vs. Number of Domains) . * . . . . . . . 29

Figure 13 Scene 3 (Domains in a* vs. b* Plane). . e .. .*. . .a. 30

Figure 14 Scene 4 (Domains In a* vs. b* Plane).. . .* . . . . . 31

Figure 15 Scene 10 (Domains In a* vs. b* Plane) . . . . .9. . . . 32

Figure 16 Scene I1 (Domains in a0 vs. b* Plane) .*. . . . . . . . 33

Figure 17 Scene 12 (Domains in a* vs. b* Plane) . . . s . . . . . 34

Figure 18 Scene 13 (Domains In a* vs. b* Plane) . . . . .. . . . 35Figure 19 Scene 14 (Domains In a* vs. b* Plane) 9 . . . . . . . . 36

Figure 20 Scene 21 (Domains In a* vs. b* Plane) . . .#. . . . . . 37

Figure 21 All Scenes (Domains In a* vs. b* Plane) . a . .. . . .. 39Figure 22 Specifying a Scene Segment. ..... . ....... 42

Figure 23 Scene 3 With Segment Outlined .... 6...... 44Figure 24 Scene 3 Segment ......... , ....... . . 45Figure 25 Scene 3 Segment, Symbol Plot, Composite .. .. .... 46

Figure 26 Scene 3 Segment, Symbol Plot, Domain #1 .. . ..... 47

Figure 27 Scene 3 Segment, Symbol Plot, Domain #2 . . ...... 48

Figure 28 Scene 3 Segment, Symbol Plot, Domain #3 . .. ..... 49

Figure 29 Example Domain Data .... . . 0...... ... 51

Figure 30 Contouring of Example Domain Data ....... . .. 51

4


8/80

LIST OF FIGURES

PAGE

Figure 31 Scene 3 Composite . . . . . 53

Figure 32 Scene 3 Composite, Symbol Plot... . ........ 54

Figure 33 Scene 3 Domain #1 ........... . ...... 55

Figure 34 Scene 3 Domain #2 . ...... . . . . . ...... 56Figure 35 Scene 3 Domain #3 ... 57........ . 57Figure 36 Scene 3 Domain #4 . . . . . . . ........... 58Figure 37 Scene 3 Domain #5 ......... . ..... . . . 59

Figure 38 Scene 13 Composite ........ . ..... . . . . 60

Figure 39 Scene 13 Composite ............ . . . . . 61

Figure 40 Scene 13 Domain #1 , . . . . . . . . .... . .... 62

Figure 41 Scene 13 Domain #2 , .... . ..... . . .... 63Figure 42 Scene 13 Domain #3 . . , .... . , . , , , . .... 64

Ii


9/80

LIST OF TABLES

PAGE

Table I Improper Clustering Parameters (No Optimization). . . 114

Table 2 Improper Clustering Parameters (After Optimizaiton) . , 15Table 3 . Proper Clustering Parameters (No Optimization). , . . . 15

Table 4 Minimum Number of Domains . . . . . . .......... 21

I

-,6


10/80

MODIFICATIONS TO

DATA ACQUISITION AND ANALYSIS FOR CAMOUFLAGEDESIGN

1. OVERVIEW

a. Background

This report presents the results of a modification to a research program 1 ' 2 ' 3

to acquire data on the spectral and spatial characteristics of natural vegetative

terrains and to develop methodologies for the analysis of these data as an aid

in the design of camouflage patterns for field clothing and large cloth shelters.The program has been concerned with camouflage for the visible spectrum and forhuman observers.

Data from vegetative terrain bachgrounds are acquired by photographic anddigitization procedures. Computer programs generate spectral reflectance curvesfor each resolution element In the scene and analyze the colorimetric character-Istics of the terrain In terms of CIE 1976 (L*a*b*) color space, Among

the outputs of the computer programs is a map showing the shape and distribu-tion of regions in the scene possessing similar colorlmetrIc characteristics.The spectral and spatial properties of these regions are to be used as the basisfor the design of three, four, or five color domain camouflage patterns.

Terrain data were acquired for vegetative terrains typical of the temperateregions of Europe and North America In both the dormant and verdant state. Data

1J. R. Goldgraben and B. Engelberg, Final Report on Data Acquisition andAnls forCamouflage sin, Decliog, Inc., Melville, NY April 191 (ecilogReport No. 236, Contract DAAK60-79-C-O072).

J. R. Goldgraben and B. Engelberg, Procedures for the Acquisition andSAnaysis of Terrain Data for Camouflage. Volume 1, Software Manual,Docilog, Inc., Melvi'le, NY, March 19 I Decilog Report 234, ContractDAAK6O-79-C-OO72).

3 J. R. Goldgraben and B. Engelberg, Procedures for the Acquisition andAnalysis of Terrain Data for Camouflage Design, Volume 2, Manual for PhotokraphicDate Acqulsition and Film Digitization, Decilog, Inc., Melville, N7, March 19 I., (Oecilog Re'port No. 235, Contract OAAK6O-79-C-OO72).

7aI


11/80

for both front lighted direct solar illumination and solar Illumination withmoderately overcast sky conditions were obtained.

This report summarizes the technical aspects In the development of modifl-

cations of the existing software for data analysis and presents recommendations

for further possible software development. Full documentation of the DataProcessing and Analysis Software and of the Photographic Data Acquisition and

Digitization Procedures are contained In references found on page 67.

The data processing and analysis software and the modifications under thisphase of the contract have been Implemented on the UNIVAC 1106 computer locatedat the Natick Research and Development Laboratories.

b. Task ReAuirements for This Phase

The following tasks were addressed under this Contract Modification:

(1) Optimization of Clusterlnj

In the previously delivered data analysis program, a histogram algorithm(HIST) was used as a first step In the clustering of the CIELAB values of thescene pixels. This algorithm Is highly efficient and greatly reduces th enumber of computations that must be performed by the Euclidean clustering

algorithm (GEOM). This histogram process Is not, however, an optimal processand the CIELAB coordinates of the final color domains can be Influenced by thecolor coordinate Increments used In the hIstogramming and by the number of

Intermediate clusters which exis t when the clustering process switches fromHIST to GEOM. The Increment levels and number of Intermediate domains are

user--specified.

it was recommended, therefore, that an optimization routine be added tothe existing color domain clustering process. The CIELAB values assigned to

the three, four, or five color domains by the existing histogram and Euclidean

algorithms are Inputted into the optimization routine which re-adjusts the

8


12/80

CIELAB centroid values using a nearest means I te ra t ive optimizat ion algori thm,

The algorithm elminates most of the bias that may have been Introduced In the

histogram clustering and thereby makes the spectral and spatial charac ter i s t ics

of the final color domains Independent of the user ' s choice of c lus te r ing

parameters.

(2) Data Separabil i ty

Each of the scenes should be analyzed and the following tasks performed:

- Develop a metric which can be used to evaluate the "goodness" of the

calculated domain centroids.

- Examine procedures for determining the "correct" number of domains

for each scene.

(3) Scene Segmentation

The Terrain Analysis Software should be modified so that the user hasthe

abi l i ty to specify any rectangular segment of the original scene asthe area

of Interest . In th i s manner some unwanted areas could be removed ( i .e ,sky,

grass) and the user could compare the lntra-scene var iabi l i ty by processing

selected segments of the scene.

(4) Domain P l o t t i n g

The symbol plo t t ing routine should have the capahl I ty of plo t t ingeach

domain individually as well as producing a combined plot. The usefulnessof

contour plott ing should be examined.

1kAlm .


13/80

2. OPTIMIZATION OF CLUSTERING

a. General Discussion

OPTIM is the clustering optimization routine. It is designed to removesome of the clustering Imperfections either Introduced by HIST or resultingfrom an Improper selection of clustering parameters by the user. The actual

algorithm for the optimization Is contained In a subroutine of OPTIM called

REDOM (See Figures I and 2).

The f i r s t time OPTIM calls REDOM, REDOM determines the domain centrold

closeso to each pixel. If the squared distance of a pixel to the closest

domali Is less than or equal to a user specified number of domain variances,REDOOMassigns the pixel to that domain. If the pixel to domain cantroid

distance exceeds this value, the pixel is eliminated from the clustering onthis and all subsequent i terations. REDOM then calculates new controlds and

variances for the domains based on the assigned pixels, calls OUTPUT, andreturns to OPTIM.

OPTIM now enters a loop governed by a user-parameter which specifies the

maximum number c.f Iterations through the loop. REDOM is called in each I terat ion

and reassigns pixels to the closest domain centrolds (which have been recom-

puted In the prior i teration) and keeps track of the number of changes In

pixel domain assignments. REDOM then recalculates domain centrolds andvariances based on the now domain assignments, calls OUTPUT, and returns toOPTIM. If the number of domain assignment changes in an Iteration, Is 0, or

if the maximum loop number Is reached, OPTIM exits its loop.

10


14/80

ENTERPTII

to call REDOM

Read LAB dis t . In variances (XVAR) beyond/

dowwhich a pixel Is not allowed In adomain for f~irst pass thru REIDOM

LOON.O,EDo0M,o

NubePixelDomai

FIGURE I SUBROUTINE OPTIM


15/80

EENTERREDOM

INITIALIZATION

v I ~crenwnt Rows

Re*ad I raw of LAD Values~nd I row of doaindn assignrnentim

UIncrement COLS

Inruint domdlnbru

CalcLwlatv pixel distancefromi dornan cunitro~dno

(~Is b) (d

aIlUR d2 TUBROTINEDOMIm n

fu1het2

l

PIx


16/80

a)(b) ()(d)

r nhtin f dom~ jiwi~No ne

an tpt isoga

EpIteumn arUat%

FIGURE a pcNixel)asind13acrul M

Yes


17/80

b. Validation of Subroutine OPTIM

The method that was chosen to validate this routine was to Intentionallyselect clustering parameters that produced Incorrect results and to see howwell OPTIM corrects for the Improper selections. Scene 10 (4 Color WoodlandCamouflage Cloth) was used as data since correct L*a*b* values are known foreach of the four domains.

[i'

The Impeoper clustering parameters were chosen by examining Table 4 of

the Final Report on Data Acquisition and Analysis for Camouflage Design 4

Case 7 was chosen and the results that were achieved, prior to running OPTIM,are repeated below.

TABLE I "IMPROPER" CLUSTERING PARAMETERS (No Optimization)

DOMAIN# # OF PIXELS L* a* b*

I 23 25.08 3.93 9.882 t0 31.99 -10.06 12.003 10 26.11 2.97 10.90S23 24.98 2.95 9.57

The above run accounted for only 6% f the total number of pixels In thedigitized scene (66 out of 1131) and only two distinct domains were produced.

OPTIM was then run (specifying a maximum of 15 loops) and the variancecut-off was sufficiently large so as not to discard any pixels on the basis ofbeing too far from an obviously bad domain centrold. OPTIM needed only 10 loops

to reach the situation where there were no changes In pixel domain assignments.The results of clustering after OPTIM are outlined below.

See Reference 1.

'I,.I.i,


18/80

TABLE 2 "IMPROPER" CLUSTERING PARAMETERS (After Optimization)

DOMAIN# # OF PIXELS L* .* b*

1 398 25.64 3.30 9.69

2 258 40.93 -3.26 16.16

3 312 30.97 -7.87 11.584 163 19367 -. 46 .85

These values can be compared to the run with good clustering parameters

with no optimization (Table 3).

TABLE 3 "PROPER" CLUSTERING PARAMETERS (No Optimization)

DOMAIN I #OF PIXELS L a*b*

1 356 24.79 3.10 9.20

2 160 41.17 -3.93 15.66

3 218 31-.S -6.05 12.104 82 18.10 0.33 -0.01

An examination of these three tables shows that even when starting from

obviously Incorrect domain centrolds, OPTIM will iteratively adjust those

centrolds until they properly represent the scene. The minor difference be-

tween the two final results (Tables 2 and 3 ) can be attributed to the fact

that the run with improper clustering parameters was forced (by the set t ing

of a very high variance cut-off) to consider every pixel In the scene.

An Interesting feeling for the workings of OPTIM can be obtained by

examining Figure 3. This Is a graph of o'b* space and shows the Initial

four improper domains and their ascnclted L* values. The centrolds of these

domains are then shown to migrate to their final values during each I tera t ion

tl OPTIM. The final L* values for the resulting centrolds are also shown.

I Ii


19/80

I,

,.rn(14e144

(i..-.,, oq) gg,

...

S / •1,,) "

./KEY

., . Cff.J. ., ., . .. .iC'"

FIGURE 3 CENTROID MIGRATION OURING ITERATIONS OF OPTIM

16

i il


20/80

3. DATA SEPARABILITY

This section describes some stat ist ical tests which were performed onthefIi L*a*b* data. The purposes.of these tests were to determine:

- For a given scene, what Is the correct number of domains?

- After clustering, how good are the obtained domains?

As will be discussed below, the f i rs t question can be answered only

as to what is the minimum number of domains which adequately represent thescene. The answer to the second question Is approached by testing whether

the domains are s ta t i s t ica l ly different from one another.

For all scenes analyzed, the minimum number of domains required was

found to be three, four, or five. For the Woodland Camouflage Cloth, four

domains were required. In all cases, at the minimum number of domains, they

represented s ta t i s t ica l ly good clustering.

a. Number of Domains Required

The Statement of Work requires that It be possible to reduce all real

world scenes tofive,

four, orthree domains. For any given scene, it seems

appropriate to ask the question How many domains should there be? 1 Phrased

differently, this problem Is finding the optimum number of domains.

Intuitively, It would seem that an analysis of the variances (squared

vector distances in L*a*b* space) of domain and pixel L*a*b* values could be

used to solve this problem. Figure 4 Is a two dimensional Plot of four

Idealized domains.

17


21/80

b*

02

D1Pl

CC

P 2 a*

DI through D4 represents Domains I through 4

C1 and C2 represents the Centrolds of Domains I and 2

P1 and P2 are two Pixels within Domain I

FIGORE 4 FOUR IDEALIZED DOMAINS

Note that all of the pixels used to cluster to the four imaginary domains

are contained within the outline of the domain. Now, the separation of the

domains should be related to the dlstaný.e between the centrolds (C, o C2,C1 to C3 , etc.) and the compactness of the domains should be related tothe distance botween pixels within a domain and the domain centrold (C to

PI, Cl to P2 0 tc.)

The very use of the terms, between and within, suggests an Analysis ofVariance where the statistical test of the "goodness" of the clustering isthe ratio of the between cluster variance to the within cluster variance.This ratio, referred to as the F ratio, can be used to test a hypothesis thatthe obtained clustering arose from chance alone. If the obtained ratio Is

sufficiently large, this hypothesis is rejected and It Is Inferred thit theclusters are, In fact, distinct.

Before proceeding In this manner, a literature survey was conducted Inthe Image Analysis area to determine what alternative procedures might have

Ia•

k4'


22/80

been developed. One paper5 was found which discussed measures of the goodness

of clustering. It discusses two useful cri ter ia which are defined as 54 and

B is equivalent to the F ratio discussed above. B is the product of the B 4 5____. between and within variances. B is said to be useful since, when each pixel

Is a domain, within variance Is zero, and when all pixels are assigned to one

domain, between variance Is zero. Hence, In both extremes B Is zero. The

authors state that It follows that the best clustering occurs at the number

of clusters which maximizes B5.

Decilog could find no logical basis for this assertion. Consider theprocesses which occur In clustering of a fixed number of pixels. If each and

every pixel has a unique L*, aft and b*, clearly the "best" characterization of

that scene Is n domains, where n Is the number of pixels. It Is Irrelevant, at

this point, that this Is of no value to camouflage design. As pixels are

assigned to domains, and provided the number of pixels per domain Is fair ly

large, the between domain variance will also Increase as the number of domainsdecreases. If a good clustering algorithm, such as OPTIM, Is used, the

within domain variance will also Increase as the number of domains Is decreased.

Therefore, B5 will monotonically Increase as the number of domains decreases and

Yll be a maximum at two domains.

B5 was calculated for each scene analyzed to date, and was found to be

a maximum at two domains, and to decrease with Increasing numbers of domains.It Is concluded that 65 Is not a good criterion. For this reason B4 (or,

equivalently, the F ratio) is suggested for use to evaluate the number of

domains required. 94 Is the ratio of the between domain variance to thewithin domain variance. This then can be a measure of the extent to which

the domains account for the total scene variance.

One simply chooses the percentage of the total variance In all of thepixels being clustered which he wants to be accounted for by the domains. In

the analysis done to date, 90% of the total variance has been the criterion.

5G. B. Coleman and H. C. Andrews, Image Segmentation by Clustering,

Proceedings of the IEEE, Vol. 67, No. 5, May 1979.

19

"F "' 'r ,,., :.• . . , :• , • :•: :' , ',• •': '•, •',',•/•,•'.* ,*• . j9 . . . . . . . . -'-',•"t


23/80

One then reads from a Table of Critical Values of F, the 90% values for I-n

degrees of freedom, where n Is the number of domains.

Figure 5 Is a plot of B4 for the Woodland Camouflage Cloth for onethrough seven domains. The dashed line Is the cr i t ical value at which thedomains account for 90% of the total variance. If one goes to the next higher

Integral number of domains from the crossover, one sees the minimum number of

domains which account for 90% of the total variance. Any higher number of

domains will account for a greater proportion of the total variance. It may

be noted In this example, the Woodland Camouflage Cloth, that four Is theminimum number of domains required to account for 90% of the variance.

Figures 6 through 12 show similar plots for other scenes which have

been analyzed. Note that the minimum number of domains is three (Scene 13)

and the maximum Is five (Scenes 3, 14, and 21; Table 4).

Figures 13 through 20 show the L*a*b* values for the domain centrolds

for each scene analyzed to date, together with a verbal description of the

scene. Figures 13, 14, and 16 are dormant scenes. Note that all centroldsare In the "red"' quadrant, In Figure 14, It Is believed that the two centroidsbelow the horizontal axis may have been caused by digitizing the sky background

between the limbs of the dormant trees.

Figure 15 depicts the domain centrolds obtained from the analysis of theWoodland Camouflage Cloth. Despite the fact that there Is no traceability

between the reflectance targets used In the photographs and NLAB's colorimeter,

one can observe the remarkable similarity between these values and those mea-

sured by NLAB's.

20


24/80

TABLE 4: MINIMUMNUMBER OF DOMAINS

SCENE # TYPE DESCRIPTION -MINIMUM-DOMAINS

3 Near Deciduous, Brush (Dormant)S

4 Far Deciduous, Brush (Dormant)14

10 Woodland Camouflage Cloth4

11 Near Fruit Orchard (Dormant)14

12 Near Deciduous, Brush (Verdant) 4

13 Far Coniferous (Verdant) 3

14 Near Deciduous, Brush (Verdant)

21 Far DecIduous, Brush (Verdant) 5

21

JJ


25/80


26/80


27/80

SCENE

30 m

t28* *'

26 .

24

22... It

20 I

16**'

10

44T I

2

, I 4 ,

~~ w mlnim4ni BETA 4

*K I FrequiredI I I @ with qptimliation,

FIGU'RE 7 Soe4 -Beta vs. Number: of Dommiins x *ithoi t optlinization

24 .


28/80

SCENE I I

30

26 .

2iI

22 J

20 .

12,16 I

lea

} 16 am

6K

II

22

=I

p :

' . . . 67 8 ,

"..: 2-~. . . . . ..... ~ BE..A....

††††††**-" .... .T".,.. .*.*€.... L

=aI ~ ~~~.. .. ..... i.. ...... •...._. .'. i........ ,, ,...... .


29/80

SCENE 12

30

28

26

24

22 I

20 f

18

16

12

6• ,2 . . .... ..•20 II.. I .. .

1 2 3 .. . 7 B 9 i0

NUMBE,0' DOMAINSiK

minimom BETA 4required(B with optimization

y wl thout optimization ,FIGURE 9 Scene 12 Bit vs. Number of ,OWeAnho ,

26:1: .. r' ''" •• ••, • ,,..J • • .•- ' •"'•I 1 II II~ I IPI


30/80

SCENE 13

30

28

26

24.....

I .,

"22 V. .

20.

18 .

tI~

1..

,

12.

* I,

,. I i.

2 '-

1 8 o

I .. . . I' . I

.K Y

Iminimom

BETA i

required

~J with optimlia~tion

6~~wh

toptimization-•.-l,

F GURE,10 Sl ane 1 .3Dta vs , umber of Domains

i

27


31/80

SCENE 14

30 I

28

26-

24 I .. .t . ,

22 I.

20 I0 .. . .. . .

18

16 *

l7

NI T Q ý I

I , , . .,

'KE'1 0 I ..

- ........ 8 '. .. . .. t . . . . . .i \1

. ........ . .* ,-.I.I , .* i ' I "

I I....II ,' • ,'i I T' • I. .. " . .

.,* , .

' N . . I t

S . . . I. .: .. . : It.-. . . ." K.E ,S. . . .. ' -- mlnlm~m BETA 1 ,

FIUE1 c n i Bet vs 1ube f with Cptnimll.tlon

, 28


32/80

SCENE 21

30

28

26-

22-, I

20 "

II1416

'I 2l . l

tO I.

N OF IS. I..... .. .

I *

'

I..u.im~~miii......... J..i.'i........ . , J.. s .. ...

minimum BETA 4

',f,, , requl red• with opt imizat ionFIGURE 12 Scene 21 Bets vs* Number o f ; l o a n0 wihuIotnzto

S~29

35


33/80

SCENE DESCRIP,.. 7.- . W- 14R DORAAT

HEAR. CLEARSl. [ .

~ *~*~4-V +14-:

I 'i - .ii 1 . , II 1 tI ,

i% (35-74).. , , I

I I,,tI n ' Il :~I ,ilii .,1 I

20 ... 2 ,I I I i i.i I' , ,*-* 4 W 14.''

2 ..3, I

1 . 1 5 2q:

, v(al1e)

FIUR 13 Scn 3 ,a f m a* v s . b* Plane

SI, . . . . 9. 9, 1'. . . . . . . . . . . .. .* .,*;** . I,

, , , ., . I .

. ...

"I ... -. . . . . .. .. . . . '" .. . . .

I ' :i ' I

.9" ,.. .; . .S. . .. . . . . . I ' ' ' '

*. I.(2 7I,). ... . . . .S. ., I *

, 9 , . 9 I

, - , * i . . ..

, i • . . .... RE..c.n.3..D.a n.. v.. ..ane, , . . . ' . , t i i ,

35


34/80

35,.

9 ii

7 ....

FIG R, SI w:

S - , I . .... . . O 9• ' 9 ,

; .

.9

. , 9 9 ,'9 * 9

'ii , , " I

'" ; .. . .. .19 .1 ,I t ,

I, --, 9 I t "9.1'

' ...C t''o. ,91,. . . : j ; 'K 2

S,~~ ~ . ., ,, :I ;t, ., , ,9 ,

, ,I . 91l t "

9. .......

'" -. 9 .

I ...... ..,

|.II

,,,I..•. .. . . . . .. . .9. 7,

' '

.. , , , :. • ' . '

I . .

, , , ... .. : , ., , , * value•

FIG, li S c n , , o aIn a* v. b, P• i,'

S__"_

t,;T. .._ '"_~. '' ~

SCENE 1035


35/80

35

SC NE DESCRIPTI,

I CLOR ýAMOUFL E O.

N M A A . AATLY LdUD .

I, ,

(41-26)T

'.... . .. . .... . . . . . .....-* '-

Fr l I I'

i ti

'~.VIi •

II I

.

. - ..'

,I,

, I I , ' ':

(I 1.62,I

U. . Ia* h* Plans

,I,32

S,( I~z= ,(,', ,

9.,

*I

, ,

K

. . . ".

,,

• ' '

t , ,. : , , I ' ''

I. : •.I=• • , .

, . . .. I

. 1: ,....I, ,

' l ,~~Iit w• , .. ..

IIGURi 15 Scen 10 Dom

ins In I VS h'I Plane•ll

Si ,, , I.$;...........-7..........


36/80

I ~ J Ilk.

7b-E , S aqI

7-I WCA~Q. O; E

It

T' I

k I1i;

I'I

,1,, 10 2 CI.0

2 qI

LN Wu

F I UE1 cn %Daen n v.b ln

* *1*33

SCENE 1235


37/80

,,.,C1)1 py f

*(1 -9HA, CL

r. 's ..

.15 5 Ia.2

L*alu

FIGUR7Soe11 Dmis na s.b in

*l I I Iji34

SCENE 13


38/80

35,

SCENE 13 ~E$CRIPT19c I

-7 -

-axi s

1~ it

,

.0 10 . 1$-20

... . I0,, value

, ' . . . . . . . . . . . . . . " l . . . i I

F U 18 S e D a In vs .* P

,35

' ' ' I. . . . l t 'I il i . . ' ' l,I. . ; I I I ,, II. , ,

' ... . . I ' ,.t', . " . . . . . . . . .

1.. . . i I•,, ..LA•

V K .i

, 15 ......... LLFIGURE-18 Scene 13 - DoIamn In * vs . -Plane

, I ,i , , , I I .. . . .i , i3


39/80

i i

S. . .... i

., *II, , CLEAL I

7I4 ( 2 5 . 9 - 9 ) 1

i t t 1 1

0(20.0Q, ) "

S. . . . . . ... . .. .I . . . . .. . . . . . '4 • " " I 'j,' - ' ' . I:I

.) , II ,I , . (1. . ...

x I q,

. .. . .. . . I,: . ,

, ,- .

10.:

I

. . . . i . .

I.

,, )iL,,,

FIGURE 19 Scene 14 - Domai In i*a. bI Plane

36


40/80

SCENE~2 OESC

I oi

.30 -7 , y

21

(15.07) * (16.30) '41

. . . . . . . .

FIGURE 20 Scene 21 - Domalns In a* vs. b* Plano37


41/80

'The remainder of the scenes are either coniferous or deciduous-verdant.

All points are in the 'green quadrant. Each has a "black" domain. Figure

17, for example, displays a pecul iar i ty of some deciduous-verdant scenes.Note the correlation among the values of the L*, a*, and b* values of the

obtained centrolds. 1he a* and b* components have been referred to by

RAmsley 6 as an Iso Hue line. In this tase, the value of L* Is also correlated

with both the a* and b* values. Whether or not this has any significance has

not been investigated.

Finally, Figure 21 Is a composite plot of the domain centroids for all

scenes analyzed. Also shown Is the space enclosed In a*, b* space by thecentrolds of the Woodland Camouflage Cloth. Since It seems Intuitively logical

that camouflage cloth should avoid extremes, visual Inspection would Indicatethat the choice of L*a*b* valies for the Woodland Camouflage cloth was good.

An interesting exercise would be to start with these centrolds, weighte.d by

the number of pixels, and cluster to four domains. A comparison could thenbe made between the camouflage cloth values and the four domains. It Is pos-

sible that, although only eight scenes have been analyzed, they are somehowrepresentative of the world on which the Woodland Camouflage cloth was based.

b. Goodness of Domains

After clustering to the minimum number of domains, It is desirable to

test the qual ty of the clustering. As will be discussed below, the stat is-tical test compares the spacing between domain centroids with the "compactness"nf the domains. The purpose of the test Is to determine If the L*a*b* valuesof the domain rentrolds arre stat ist ical ly signif icant ly different. If theyare, clustering is good because the spacing between centroids is large, and

the domains are 'compact . The stat ist ical tests will Indicate the probabilitythat the difference in LOa'bic values of any two centroids occurred by chance,

as opposed to the difference having arisen from real characteristics of the

scene. The results will be discussed below.

6 A. 0. Ramsley, Selection of Standard Colors for the Woodland CamouflagePattern , US Army Natick Research and Development Laboratories, September 1981,(Technical Report NATICK-81/030).

38


42/80


43/80

Prior to this discussion, however, the practical significance of these

tests should be evaluated. Due, at least In part, to the fact that very large

numbers of pixels are assigned to each domain, the stat is t ical tests become very

sensitive. As a result, very small differences in the location of centroids

will be found to be "stat ist ically different . Thus, If a scene has been clus-tared down to ten domains, for example, the stat is t ical tes ts could l ikely

Indicate that there are ten significantly different domains at the confidence

level chosen.

The actual distance between some pairs of centrolds In L*a*b* space may

be small, even less than I unit, which Is the theoretical visual difference

threshold. Theoretical is underscored since it must be remembered that It Is

based on the 1931 standard observer, using the method of paired comparisons.

The threshold which Is applicable to camouflage cloth when viewed under field

conditions Is not known, but Is probably larger than I unit.

Therefore, It should be kept In mind that the term significantly different

or different refers to the precision and separablllty of domains from a sta-

t is t ical viewpoint and not to the visual difference between domains as applied

to camouflage cloth In countersurvelliance.

Student t and Fischer F tests were run to test the stat is t ical significanceof the distance between centroids InCIELAB space. The F test considers all

domains simultaneously and compares their combined variance with the spectral

veriance In the overall scene. The t test compares distances between Individualpairs of centrolds with the pixel variance within the respective domains.

Both the t and F tes ts are calculated and printed out in subroutine OUTPUT

for all clusterings of ten and fewer domains. Because the t tes t operates on

pairs of centroids, multiple t tests had to be performed; and In order to keep

the confidence level within each test high, an-a priori confidence level of

0.99 was chosen. This means that If the obtained value of t exceeds the cri t i -cal value or the probabi l i tyls 0.01 or less, the difference Is due to change.

As It turned out, for the scenes analyzed to date, the values of all t s for

140

'. . l . l.. . . l'l1 l . .. .


44/80

ten and fewer domains are far greater than the cri t ical value for 0.99 confl-

dence. It Is probable that, In the future, even higher confidence levels can

be used.

On the basis of the t tests it Is concluded that, with as many as tendomains, each and every domain Is different from every other domain. Given

this separability, It Is Inferred that the clustering is good .

Because the F test compares each domain with the distance to all othercentroids s$inultaneously, a lower confidence level, namely 0.90 was chosena priori. The results of the F tests confirmed the results of the t tests.

Again, with as many as ten and as few as two domains, each domain was compact

as compared to the separation between all domains, Based on this conclusion,

It Is again Inferred that the clustering Is good ,

The obtained F and t s ta t i s t ics for each run analyzed are shown In

the Appendix.

41


45/80

4. SCENE SEGMENTATION

a. General C~scusslon

One of the extensions to the software for the Analysis of Terrain Datafor Camouflage Design was the abil i ty to process only part of a total scene.

In the original software package the user specified the number of columns and

rows of the entire scene and the ent i re area was processed. The user Is s t i l lrequired to specify the number of columns and rows of the ent i re scene but alsoto specify that part of the scene that Is actually to be processed.

As can be seen from Figure 22 the user Is required to specify the lower

left hand and upper right hand corner of the rectangle of Interest. This is

accomplished by setting the variables ISEGRI and ISEGR2 to the starting and

ending r&A numbers, respectively. The variables ISEGCI and ISEGC2 are set to

the starting and ending column numbers, respectively. The Terrain AnalysisSoftware then works on this scene segment in exactly the same manner as If it

were the entire scene.

I ISEGC24) Row IRODWX

Column ICOLMX

ISEGR2 -

SISEGRI

Row I _Column I

iISEGCI

FIGURE 22 SPECIFYING A SCENE SEGMENT

42

S•'' .. •.. .......' ....... .•..:.::• :,"....... •,..:•.:• •' . . . . • ••°- • '• ',,;•,,:,• • , ,• 'v.,,,.: ., ,. ., .-,... .... , ...... .. ..... .. Ar


46/80

b. Symbol Plots of Scene Segments

The modifications that were nectissary to Incorporate the abi l i ty to

process scene segments extend to the plotting software. It seemed reasonable

that the entire scene should be outlined and the scene segment plotted at the

proper offsets. This allows easy comparisons of multiple runs of the same scene

as well as a quick visual check of the area of Interest. In order to accomplish

this It was necessary to pass the coordinates of the segment, In addition to

the size of the total scene to the plotting routine (SYMPLOT).

An example of this appears on the following pages. Figure 23 Is a plot

of the entirety of Scene 3 after being clustered to 3 domains. Arrows havebeen hand drawn to Indicate the scene segment that was specified on a subse-quent run of the analysis software. The scene segment coordinates that were

specified were ISEQRI-30, ISEGR2460, ISEACl=5g, and ISEeC2-80. The resultingplot can be seen In Figures 24 and 25.

The scene segment that appears in these figures was not generated by clustering.The option of specifying centroldal L*a*b* values and then assigning pixels

to the closest domain centrold was used. This was done so that the segmentplot and the segment area of the entire scene would be identicAl. It Is

,i important to realize that If we had clustered the scene segment the resultingcentrolds might have been very different.

43

I'I. .

rr


47/80

rr

I. A:

.9 4

FIGURE 23 SCENE 3

L44


48/80

III ' -( .L F'o m U(6ALV fi~CTOR f0BJ CM/PLOI CW 2-21

100 .40* 104.90, 104 .4O. 100.00, -3L ?7U 'JU.O)()51,60, 83.70# 90-20, 78.30. 714,0,

W4IN -3 N'ýHT(.H = INCR =I

IAH0tdt -I INTE.R'V =2 ISEQCI ;:501&1 V 9

Ni WCMX 31 N V R X 1 IGE.GRt=30 1 0 K",t 40 0

DOJM IN NUM1BER2 3

bNMO~.+x

L J* )1 l~ 40 2 .49

F4 5 4 43.68

ki ' ii 1 rip 4.64

FIGURE 24 SCENE3

45


49/80

F I U E P 1 C N E H N T D C H O I E OAN

-- .---. -. ~-.. - -- ~- ________________

' .. .... . ......,.................. ....... -'..... ...... .. . -- --. ... ... •


50/80


51/80

II

FIGURE 27 SCENE 3 SEGMENTED, DOMAIN 2

48


52/80

FIOURE28 SCEE 3 94KENTED, DOWAN3

ii9


53/80

5. DOMAIN PLOTTINQ

a. Individual Domain Symbol Plots

The composite symbol plot that resulted from an analysis of Icene 3 I(Dormant, Deciduous Brush-Near) is shown In Figures 31 and 32. In order to

give the user a better appreciation of the shapes associated with each of thedomains, the software was modified to allow for the plotting of each domain

separately. The Individual domain plots for Scene 3 can be found In Figures33 through 37 and for Scene 13 in Figures 38 and 42.

The user has the option of plotting the composite alone, the Individualdomains alone, or both the composite and Individual domains. The capabilitycan be combined with scene segmentation end example results for Scene 3 areshown In Figures 24 through 28.

b. Contour Plotting

The usefulness of plotting contours as compared to the supplied symbolplots of domains was examined. It Is felt that contour plotting of the compos-Ite domain data would not be of any help In the graphical representation ofdomain shapes. These contour plots would be arbitrarily misleading. It shouldbe pointed out that contour plotting Is not the same as shape outlining andthis discrepancy Is what creates most of the problems. A contour plot would

graphically show the slope, or rate of change, between domains. This wouldarbitrarily assign a greater weight to a change from Domain I to Domain 5 thenfrom Domain I to Domain 2.

Even if the contouring routines were used on the composite date to outlineone domain at a time, similar to the individual symbol plots, the results would

50& ,I


54/80

be misleading. Consider Figure 29 as a set of pixels with their associated

domain numbers.

~1222 1 1 1 33311 222 I 11 3 3 1S1222 1 11 3 3 3 1

1 2 2 1 1 1 3 3 1

FIGURE 291 EXAMPLEDOMAIN DATA

If we were to draw contour lines at levels 1.5 and 2.5, In order to

outline Dom In , we would generate the plot shown In Figure 5-2,

1:, - 2 2 1 1 3 3 111 2 21 13 3 1

1 21 2

, ~I 1 1 1 1-- ---

FIGURE 30, CONTOURINGOF EXAMPLEDOMAINDATA

The contour lines that were handdrawn were squared off for simplicity.

The resulting plot, however, Illustrates the problem with contouring. By

definition contouring assumes that there must be a Level 2 between the Level

1 and 3 areas. This Is Indicating a faster rise on the right-hand side as

compared to the left.

A method that might be of some use Involves modifying the data file that

Is contoured, If the data file Is edited so that only one particular domain

Is left Intact, and all other values are changed to a background value, then

contouring would outline this single domain. 'In this manner, by submitting

five versions (each of which has only a single domain and a background), of the

original data file to the contouring software, outlines of each of the five

Individual domains could be produced.

i 51

. . . . I-. -

-,i' t ,


55/80

The Calcomp Contouring Package available on the UNIVAC 1106 has a size

limitation that precludes using it on a complete scene. It would probably bemore efficient to write shape outlining software than to attempt to expand the

contouring package. Shape outllnlng software would not have the false areaproblems of contouring discussed above.

52

d.ig


56/80

RANGE 60 M SCAILE FACTOR (I3BJ CM/PLOT CM) 9.42

11LUri H/CM'-MICRONk FjO.00, 82-8O,03.50. 1049000 117.80.11~~0

87~.60. 58'.70, el.Z0, 78.30. 71,60,

NFN:5N'WTCH r201NCR 2

ITHRSM 324 INfl.RV = 6I ~ 2 ~ 2

NEW4CM'i= 120 NEWRMX=132ISGORI =I ISE.0R2 =132

DOMAIN NUMBERC..

123 4bGYMOOL + Y

L3b-74 21717 23.1022.13 17 .f36

5 28 1.0 .33 1.6345

817 72 11.90 0.39 5.841.64

F GURE 31 SCENE 3 CONOSITE

53


57/80

V V

FIGURE 32 SCENE 3 COMPOSITE

.5i4


58/80


59/80

++

.9 ,*,',.,, 4 I" I { •. ,)*'t- ,1

- +' ••.- - •+ . 9.• ' ,• .1.3 -* • :. • . ++ i + ' .:

. . ., • .9. I " I

4. +,.).

-r : .t" * ,. .9.•+

1, W,

* I.. $" .

l .*..+

++ '+ .). $4++ +l•'::.

+ + 14+

+4., ,

+ +

44

+

'

+ 44

4+ +9 -.+4

744r A. A 1069

FIGURE.9 34SEE3.OAN#

~4 9 ~9.45*


60/80

n xn X.~ rr N M fK 2Rx¾XN K XAlO R0l(KNx w

*,& )*

xK K

Ni(x xxx

x~~~~K NMxA

Nt )47X SN K NbMW N* NKK N N)A

irx x x

N

x x xx - j X~ oxNxw

N x N

FIUR 35 SCN 3 DOAI 3x 1%1 X x~ex x lb7

________ ______.."___


61/80

JII

ON

x xxx ~ x_

xK

IL ~ K\~ Kxx

L, 04 )

10 X IM NK~ )k

xK

"FIUR 36 SCN 3 DOAI X~

x I58


62/80

Is

vii f

Smom

FIGURE 37 SCENE 3 001, AIN #

'I59


63/80

Cl"

C-4

0f ;C; AU

C; N

, -4,~

( C ; la (b) C

0 - CD

C) 1*0

CC t ~ItIt

o A 0 N

0j 0

z 0

0 X

600


64/80

4,1

61


65/80


66/80

I.L

A til.

I.' ~, 63


67/80

01..

rrt

A. 1InN4.ILb

:1 IL .

ii

I'


68/80

6. COMMENTSAND RECOMMEhrJATIONS

a. Scene Diqitization

It is strongly recommended that additional scenes be processed and

entered into the available scene date base. This would dramatically increase

the uti l izat ion of the current analysis software.

b. Interactive Data Analysis with Graphic Output

The current data analysis software was written for the NLABS UNIVAC 1106

and operates In a batch processing mods. Although this was a reasonable and

cost-effective approach for the Ini t ia l development phase, the needs of the

camouflage designer can best be served In an Interactive environment with Oraphic

out.put'

The current analysis software was structured In ouch a manner that itcould be transferred to a mini/micro computing system (with a 16 bit word length)

with only minor changes. The PDP 11/23 computer which Is serving as a controller

for the spectrophotometer would be an ideal host system for the Interactive

data analysis software. The difference In the word lengths between the

UNIVAC (36 hits) and the PDP 11/23 (16 bits) has been analyzed and determined

to be no problem.

An Interactive menu-driven executive would need to be written. This

executive would be responsible for guiding a user through a data analysis

session as well as being an interface between the user and the graphic output

devices.

Alternate output methods should be analyzed to Increase the utility of

the resulting domain assignmont plots. These alternate methods should includevideo display systems (both Black and White and Color) with hard-copy capability

as well as additional pen plotter presentations.

65

U...................................ii.~.... .. Ii"I1'...Ii ilII i" i..... i


69/80

c. Alternate Data Collection Methods

The current photographlc/digltlzatlon procedure Is time consuming, and

with respect to some details, less than optimum, An investigation of alterna-

tive techniques should be performed and a procurement specification should be

deve loped.

As a minimum the approaches that should be examined include the acquisit ion

of a black and white television camera, solid state array cameras, and solid

state cameras with spectral outputs,

Systems should be analyzed for portability, re l iabi l i ty, camera sensitivity,

resolution, signal to noise ratio through the Interference f i l ters , and other

parameters st i l l to be determined. The goal should be to provide a specificationforacommercially available off-the-shelf system; however, special puv'posedesigns should be examined for completeness.

This document report$ research undertaken atthe US Army Natick Research and Develop-ment Command and has been assigned No.NATICK/TR-.•./QA(_ in the series of re-ports approved for publication.

66

... .. .. .. . .. . .


70/80

REFERENCES

Coleman, G, B. and H. C. Andrews, Image segmentation by clustering, Proceedings

of the IEEE, Vol. 67, No. 5, May 1979.

Goildraben, J. R. and B. Engelberg, Final Report on Data Acquisition and

Analysls for Camouflage Design, Decilog, Inc., Melville, NY, April 1981

(Decllog Report No. 236, Contract DAAK60-79-C-0072).

Goidgraben, J. R. and B. Engelberg, Procedures for the Acquisition and

AnalysIs of Terrain, Data for Camouflage Design, Volume L, Software.Manual,

Decilog, Inc., Melville, NY, March 1981 (Doecilog Report No. 234, Contract

DAAK60-79-C-0072).

Aoidgraben, J. R. and B. Engelberg, Procedures for the Acquisition and

.Analysis ofTerrol t Data for.Camouflage Design, Volume II Manual forPhotographic Data Acpulsitiorn and Film Digitization, Decilog, Inc., Melville,

NY, March 1981 (Dacllolj Report No. 235, Contract DAAK60-79-C-0072).

Ramsley, A. 0., Selection of Standard Colors for the Woodland Camouflage

Pattern, US Army Natick R&D Laboratories, IPL, September 1981 (Technical

Report NATICK/TR-81/030).

67

|,, -- . . .


71/80

(This page intentionally left Blank)

6B

':' .. . . .,.• . • , . , , 'mt .• •.••,•,•,•• •j ' , • :• .• , •'., ,'• • ,'•.••,,• j•,,•, • ,.J, .,• • ,• €.•.• • •- , • , , • , -I


72/80

APPENDIX

Outputs of Terrain Analysis Software

Showing Statistical Date

69


73/80


74/80

WN

Ai 400#

ar ID U 0-O44

a itp'v4A' 0.4

vat l.0f

6 or N~*,3V1-c% 44

04Nodc

VACSA N O O4. *Af.1'MmC 044 T o .

9-@"ef0 3 5

V16 I JK% d bm 'a-A CE#.t a

4 ~ ~ wt .0 W snJ S-I@ enhW a- f. -wi-fwN)@SJ~ w9' A vi PO M-

mi 44, 4,-M gV 0 Sa 5n%

iej1 1-4r .Sd~5 W- I Ir~.

L iU4 a-


75/80

Clb

* In

.d .4

qoI I 0NNM' 0WNNWOMO =

Ity w

~~~fWE"jI- 72


76/80

Itt

1.

M

.0 1 1,r%

It,

4 44

Al 4%j a W.' ,' , Vttts'.,gpe

4u~;. +

1144

le~~~O P01on nJ90 'P I I~iS'i~I1

01 0 1

s4 lldý in)" a . 4 6

I .. -l.' 4 a. 0 a 0 0 0 0 0 .:

AS IA ti

t'hN)4WiSIC ENE IC#PNON 3 4 73


77/80

w

NNIin

itp

M H n o Cq

U - ~

4m4

S t.

rI

P4..4 St 4

to ,i V 11 t --%D I C t 9-14'f9 I Q dt .lif 0 .S

4-

all Il 11I. t' ~ .414 N J' fI l

, 9, e 4

Ifo* 11. 1 ;,I P44 V C.V4 1c 0V

WI W aOW It:- f 1 t of1 , 1- 4 lit*

"Of. k, C. cl a

a' 0' 1TJ &Il ,BA L '0 4.' 1-( k 1

4-, w I,*V* r o-1 I 'UVCC .0 L .0*1 . '1 C '4L a so - *%t 4 .tf h* t' '4~P4B44'jlI~~~~~~CN 120 SSOSSSSSSSS*SU

~~ 4) 4~~~9~ ~ 4 9 -- IJ.9. t. *I *U(4 4v5N BW


78/80

. . . . . . . rm fl...rv .


79/80

.00

* 'I

* I*

'A 4

0 Ia- .,

-cN

44 "vI I-"If

% t.fUlrwd Oi %G0AaA1" 0-0 W

0. 0JqNN A

II 00M N b

10Ims4A NO

NI-0v- 000 - @ *U... U 0#0 %4 .2

SCEN 14xiv- IN


80/80

6*%

epic

in&I

10 1 si N

SI" M *1400Ol 6 c K.11- - N~ wP _ 10t W#0 W1uI' W%#G CQUCdit moo*-~

-to m4 40

U-40~~.AV3O't!2

9 W%-c-'P 9U

to It iflo -f0h~ 04

j og It:-h -- hMA fW%% P4 4c 4%11w4

ace*

SO N 0 2 * ( * *0 0 0 0100 0 6 f o d0 8 0SSO *

Date post:	03-Jun-2018
Category:	Documents
Upload:	sawankuya
View:	243 times
Download:	1 times

CAMOUFLAGE DESIGN

Documents