ROT ^

Km RESEARCH REFERENCE CfllLECTIO' f: @S NATIONAL CENTER, MS-521

UNITED STATES i""' DEPARTMENT OF THE INTERIOR

GEOLOGICAL SURVEY

22002

SUMMARY TABLES FOR SELECTED DIGITAL IMAGE PROCESSING SYSTEMS

Open-File Report 77-414

UNITED STATESDEPARTMENT OF THE INTERIOR

GEOLOGICAL SURVEY

SUMMARY TABLES FOR SELECTED DIGITAL IMAGE PROCESSING SYSTEMS

By Virginia Qarter, U.S. Geological SurveyFrederic fill ingsley, Get Propulsion Laboratory Jeannine Lamar, Rand Corporation

Open-File Report 77-414

mw

1977

CONTENTS

Page

Abstract 1

Introduction 1

System Summaries 3

G.E. Image-100 3

Bendix MDAS 4

I 2S System 101 4

ESL Idims 5

Comtal Series 9 5

ISI System 470 and Sand 6

Kandidats 6

JPL Vicar 7

Purdue Larsys 7

Tables 9

Definitions 36

General purpose image processing 36

Radiometric corrections 36

Digital level slicing 37

Contrast stretch 37

Geometric corrections 37

Geometric warping 37

Registration 38

Page

Geometric interpolation techniques 38

Filtering 38

Transforms 39

Multispectral analysis 39

Supervised classifier 39

Unsupervised classifier 40

Mensuration 41

Inter-band manipulations 41

Statistical operations 41

Inputs 42

TV camera 42

Microdensitometer 42

Digital tapes 42

Analog tapes 43

Outputs 43

Volatile outputs 43

Hard copy outputs 43

Other features 43

Microprogrammable processor 44

Present computer 44

Software 44

n

TABLES

Page Table 1. General purpose image processing 9

2. Multispectral classification 18

3. System inputs 24

4. System outputs 27

5. Other system features 33

SUMMARY TABLES FOR SELECTED DIGITAL IMAGE PROCESSING SYSTEMS

By

Virginia Carter Frederic Billingsley Jeannine Lamar

ABSTRACT

The exploration of earth from space has greatly increased the demand for

computer systems of great speed and complexity.

Because it has proved difficult to make comparisons between systems in an

orderly fashion, a set of five tables have been developed for the comparison

of digital image processing systems. The tables have been compiled for six

U.S. manufactured hardware/software systems. Three U.S. software systems

are included in the tables also.

The tables and the System Summaries have been prepared primarily from in­

formation supplied by the manufacturers. While some rewording has been

done to make the descriptions parallel, the manufacturers' wording has been

retained to a large extent.

INTRODUCTION

The exploration of Earth from space has greatly increased the demand for

computer systems of great speed and complexity. Among the many functions

of these computer systems is the processing of digital images such as those

produced by Landsat (formerly Earth Resources Xecnn°l°9y Satellite - ERTS).

There has been a proliferation of digital image processing systems, both

hardware/software and software, to analyze these data.

Because it has proved difficult to make comparisons between systems in an

orderly fashion, a set of five tables have been developed for the comparison

of digital image processing systems. The tables have been compiled for six

U.S. manufactured hardware/software systems: 1) the General Electric (GE)

IMAGE-100, 2) the Bendix Multispectral Data Analysis System (MDAS), 3) the

Electronics Systems Laboratory (ESL) Interactive Digital Image Manipulation

System (IDIMS), 4) the Comtal Corporation (CC) Series 9 System, 5) the

International Imaging Systems, Inc. (I 2S) System 101, and 6) the Interpreta­

tion Systems Incorporated (ISI) System 470.

Three U.S. software systems are i

Propulsion Laboratory (JPL) Video

2) the Laboratory for Applications of

3) the University of Kansas KANDIDATS

rapidly, and other systems may be add

included in the tables also: 1) the Jet

Communication and Retrieval (VICAR) System,

Remote Sensing (LARS) System, and

System. The tables can be updated

5d using the same tabular format.

Isolated pieces of hardware performing

single purpose systems where processing

are not included.

These tables are intended for system

There are many existing digital image

different in concept or operation,

these tables, nor can there ever be,

a system for a particular use. The

subjective judgements on the systems,

to take into account such parameters

only part of a system function and all

is only done as a display function

comparison and not systems evaluation,

processing systems; each is slightly

There is not sufficient information in

to make them the sole basis for choosing

authors have deliberately avoided making

The selection of a system will have

as operating speed (especially for the

multispectral classification, if that is a prime use of the equipment),

the flexibility with which other operations are possible, the ease of

communication with the system during operations, the ease of writing and

implementing new functions, and the cost of the system.

Finally, the tables have been designed for use by analysts with some degree

of familiarity with image processing. They cannot serve as a training

course in image processing. The reader desiring additional information*

is encouraged to refer to the extensive literature on the subject. For

clarification, some terms used herein are defined following the tables.

The tables and the System Summaries have been prepared primarily from

information supplied by the manufacturers. While some rewording has been

done to make the descriptions parallel, the manufacturers' wording has

been retained to a large extent.

SYSTEM SUMMARIES

G.E. IMAGE-100

An interactive image analysis system designed for multispectral classification

of satellite and aircraft multispectral scanner data, with limited general-

purpose processing available. The hardware system includes a PDPll-class

computer, an interactive image analysis console containing a hardware

classifier and other specialized processing functions, five channels of

memory for image processing and display refresh with thematic overlay capa­

bility, and a color display.

Sold as a complete system with proprietary software. Can be expanded and(or)

reconfigured to meet specialized user requirements.

General Electric CompanySpace DivisionP.O. Box 2500Daytona Beach, Florida 32015(904) 258-2924Contact: Robert W. Towles

Manager, Image Processing Applications

BENDIX MDAS

A combination hardware-software system designed primarily for multispectral

classification of satellite and aircraft multispectral scanner data, with

limited general purpose software processing available. The hardware system

includes a PDP-11 and an interactive image analysis console containing a

hardware classifier and some specialized processing functions. The system

can process up to 16 channels of image data simultaneously and has a solid-

state refreshed 3-color display with graphic overlay.

Sold as a complete system with proprietary software. Can be expanded to

meet user requirements.

The Bendix CorporationAerospace Systems Division3621 S. State RoadAnn Arbor, Michigan 48107(313) 665-7766Contact: Mr. Charles L. Wilson

Manager, Earth Resources Applications

I 2S SYSTEM 101

An interactive multi-user digital image processing system including multi-

spectral analysis and an expandable set of general purpose algorithms. The

hardware consists of a Hewlett-Packard 3000 Series II Computer with disc

image storage and user console containing up to 14 channels of refresh

memory, with hardware processing at video rate, a programmable floating

point image array processor and color displays.

Sold as a specific complete system with proprietary software, configured

to user requirements.

International Imaging Systems Division of Stanford Technology Corporation 650 North Mary Avenue Sunnyvale, California 94086 (408) 737-0200Contact: Mr. Paul Papathakis

Director of Marketing

ESL IDIMS

A combination hardware-software system for general purpose interactive,

multi-user image processing including multispectral analysis and classifi­

cation. Hardware includes: Hewlett-Packard 3000 Series 11 computer,

interactive processing console including a programmable array processor,

3 or more channels of refresh memory with video color display; refresh

memory is CCD or random access. Software can also process collateral data.

Sold as a complete system with proprietary software with configuration

tailored to user needs.

Electromagnetic Systems Laboratories, Inc.495 Java DriveSunnyvale, California 94086(408) 734-2244Contact: Dr. James E. Burke

Manager, Image Data Systems

COMTAL SERIES 9

A complete stand-alone system designed for both multispectral data classifi

cation and general purpose image analysis and enhancement of up to four

separate image planes simultaneously. It incorporates a combination of

hardware and software functions to provide maximum processing speed and

capability while maintaining complete functional flexibility. Changes in

algorithms and processing techniques may be implemented by software changes.

Major subsystems are magnetic tapes, computer, and 8000 Series display.

Comtal Corporation169 North Hal steadPasadena, California 91107(213) 793-2134Contact: Dr. John Tahl, President

ISI SYSTEM 470 AND SAND

Combination hardware-software systems for general purpose image processing

and multispectral analysis. Swing (proprietary) software and digital computer

and peripherals are the same in both; 470 contains real-time video processing

and 600 image video disc store; SAND is RAM (Random Access Memory)-based for

interactive computer graphics applications.

Specific configurations may be tailored to user needs.

Interpretation Systems IncorporatedP.O. Box 1007Lawrence, Kansas 66044(913) 842-5678Contact: Mr. Jerry D. Lent

Director, International Marketing

KANDIDATS

KANDIDATS is a digital image processing system that interacts with the usere=

at the command string level. It includes prompting for parameter input and

checks user input for errors. Image analysis programs available in KANDIDATS

consist of utility functions, image transform operations, spatial clustering,

and Bayesian classification. Programming structure and file structure are

modular.

Dr. Robert Haralick Space Technology Center University of Kansas Lawrence, Kansas 66045 (913) 864-3542

JPL VICAR

A Fortran-based expandable general purpose image processing software system.

It includes a wide variety of general purpose algorithms including multi-

spectral analysis and computation. Implemented at JPL initially on IBM-360/

44; recently modified to 360/65. Computer peripherals serviced by the

system include: 2 Comtal interactive consoles, a Ramtek refresh memory, a

quick look Polaroid hardcopy output, and off-line film recording.

Software system available from COSMIC includes documentation and listings.

Jet Propulsion Laboratory 4800 Oak Grove Drive Pasadena, California 91103 (213) 354-5016 Contact: Mr. William B. Green

Section Manager, Science Data Analysis Section

PURDUE LARSYS

Combination hardware-software image processing facility at the Laboratory

for Applications of Remote Sensing (LARS) in West Lafayette, Indiana.

System is used locally and also supports remote terminals via phone line.

The software system (LARSYS) is primarily designed for multispectral

classification, but also includes limited general purpose image processing.

Hardware is an IBM-360/67 plus an interactive console containing a black-

and-white video display with facilities for black-and-white and color

photographic output.

System hardware is not for sale, but several terminals are available

throughout the country and at West Lafayette. System software is available

through COSMIC.

Purdue UniversityLaboratory for Applications of Remote Sensing (LARS)Purdue Industrial Research Park1220 Potter DriveWest Lafayette, Indiana 47906(317) 749-2052Contact: Mr. Terry L. Phillips

Deputy Director, System Services

8

TABLE

1GE

NERA

L PURPOSE

IMAG

E PROCESSING

Page

1

of

3

RADIOMETRIC

FUNCTION

GE IM

AGE

100

BENDIX M

DAS

I2S

SYST

EM 10*1

RADI

OMET

RIC

CORR

ECTI

ONLandsat

deba

ndin

g, correction

for

6-se

nsor

unb

alan

ce o

f MSS

Corr

ecti

on o

f photographic d

ata

Othe

r

Line

by

line

ga

in a

nd o

ffset

(Opt

iona

l so

ftwa

re,

or so

ft­

ware

/har

dwar

e hy

brid

)Average

adja

cent

li

nes

NoAv

erag

e adjacent li

nes

Hard

ware

shading

(for

TV

camera in

put)

Shad

ing,

when u

sing

vidicon

option

D-log

E correction

Shad

ing

Focu

s correction

Motion d

e-bl

urri

ng

Video

averaging

soft

ware

To

r no

ise

remo

val

Atmo

sphe

ric

correction

by

regression analysis

Sun

angl

e co

rrec

tion

DIGI

TAL

LEVE

L SL

ICIN

GContinuum

B/W

to m

ultiple

disc

rete

grey shades

Assignment of colors

to

disc

rete

grey shades

8-le

vel

(to

them

e tr

acks

) N-

leve

l (O

ptio

nal)

co

mple

tely

flexible spacing

* 2-

1-le

vel:

Of

fset

, counts/level

0 offset,

gain

(8 levels/band,

equal

size)

A N-lcvcl

with fl

exib

le spacing

by card input

N-level

Comp

lete

ly fl

exib

le spacing

Line

ar,

high and

low

limi

t,

0 of st

eps

64 p

reset

colo

rs,

8 at

a time

N-level

- arbitrary

color

assi

gnme

nts

* 24 preset colors

0 Ar

bitr

ary

colo

r as

signme

nt,

A 51

2 co

lors

available

(false

color

mode)

Arbitrary

colo

r assignment,

512

colors available

Single level

color

slice

on

B/W

image

CONT

RAST S

TRET

CHTa

ble

look

up implementation

Line

al-

Equa

l occupancy

Square r

oot

Piecewise

line

ar (O

pt.)

Table

look

up im

plem

entati

on

Equal

occupancy

Square r

oot

Piecewise

line

ar (O

pt.)

Variable ga

in a

nd o

ffset

(pla

nned

)

Linear

Equa

l oc

cupa

ncy

Logarithmic

Piecewise

line

arExponential

TABL

I;

1Gi'.NLRAL PURPOSE IM

AGE

PROCESSING

Page

1 of

3

(Continued)

RADI

OMET

RIC

FUNCTION

SI, HUMS

COMT

AI,

STHI

IiS

ISll SYSTKM 4?0 ,

SAND

RADIOMFTRIC CORRECTION

Lundsat

dcbn

ndin

g, correction

for

6-se

nsor

unbalance of

MSS

Corr

ecti

on o

f ph

otog

raph

ic d

ata

Other

Average

adja

cent

li

nos

Notc

hed

filt

erin

gNo

Shading

correction

Focus

correction

Motion d

e-bl

urri

ng

No

Atmospheric

corr

ecti

ons

(planned)

Sun

angle

corr

ecti

on

Replace

adja

cent

line

Average

adja

cent

lines

Shad

ing

correction

Focu

s co

rrec

tion

(planned)

DIGI

TAL

LEVEL

SLICING

Cont

inuum

B/W

to m

ulti

ple

discrete g

rey

shades

Assi

gnme

nt o

f co

lors

to

discrete g

rey

shades

N-le

vel

Comp

lete

ly f

lexible

spac

ing

N-le

vel

Arbitrary

color

assi

gnme

nt,

4096

col

ors

avai

labl

e, 64

at

a

time

fo

r display

Colo

r contouring

8-le

vel

(har

dwar

e re

al-t

ime)

N-

leve

l software

Comp

lete

ly f

lexi

ble

spac

ing

Arbitrary

level

setting

(color c

ontouring

- SAND)

CONT

RAST

STRETCH

15 m

ajor types

of i

nten

sity

transformations.

Variety of

int

erac

tive

tra

nsfo

rms

Table

look

-up

Line

ar

Loga

rith

mic

Piec

ewis

e linear

TARL

C 1

GP.NF.RAL

PURP

OS1-

TMACli PROCF.SS1NG

(SOFTWARE

SY5T

F.MS

\of

3

(Con

tinu

ed-)

RADI

OMET

RIC

FUNCTION

RADI

OMET

RIC

CORRECTION

Landsat

dchanding, co

rrec

tion

for

6-sensor u

nbal

ance

of MSS

Correction o

f ph

otog

raph

ic d

ata

Othe

r

KAND

TDAT

S

Aver

age

adja

cent

li

nes

No Point

corr

ecti

on

JIM,

VIC

AR

*Average adjacent li

nes

Convolution

filt

erin

g Li

ne by li

ne ga

in and

offs

et

Four

ier

fiIt

erin

g

Shad

ing

P-log

E correction

Focu

s co

rrec

tion

Aver

age

several

images for

noise

correction

Individual pi

xel

corr

ecti

on

LARS

YS

Gain

an

d of

fset

*Avi

M*ag

c adjacent li

nes

*Fou

ricr

filtering

*Sca

nncr

an

gle

correction

*Convolutional

t'i Itering

*Sha

d in

g

*Focus correction

*Fin

itc

aperture co

rrec

tion

DIGI

TAL

LEVE

L SL

ICIN

GCo

ntinuum

B/W

to m

ulti

ple

discrete gr

ey sh

ades

Assi

gnme

nt of

colors

to

disc

rete g

rey

shad

es

N-level, flexible

Grey

sca

le li

ne p

rint

erN-

leve

l flexible tabl

e Li

near

-arb

itra

ry endpoints

Log

Exponential

Grey

scale

line

printer

N-le

vel

flexible table

Grey scale

line p

rinter

Arbitrary, in

dis

play

memory,

10-c

olor

*Arbitrary (a

naly

st-g

ener

ated

ta

bles

in color

space)

*Arb

itra

ry (a

naly

st-g

ener

ated

color

tabl

es)

CONT

RAST STR

ETCH

* No

t in

COSMIC

Linear

Equa

l occupancy

General

non-linear

Tabl

e lo

okup

im

plem

entation

Linear,

arbi

trar

y en

d points

F.qu

al occupancy

Contouring

Flex

ible

table

Logarithmic

Expo

nent

ial

Piec

ewis

e li

near

Flexible table

lookup

Equa

l oc

cupa

ncy

TAIW

, 1

Miur

osi;

IMAM

I'UO

CP.SS

INC,

Page

2

of

3

GEOM

ETRI

C CO

RREC

TION

5

REG.

ISTR

ATIO

N!r,n

IMA

GE 1

00BHNDIX M

DAS

I2S SY

STEM

10

1

CORR

ECTI

ONS

Inte

rnal

sensor d

isto

rtio

nsMirror scan ve

loci

ty p

rofile

(Opt,)

Roll

, pi

tch,

ya

w (Opt,)

Ephe

meri

s (altitude

and

velo

city

) (Opt.)

Skew

As

pect

ratio

and

scaling

Std.

: By line selection

Opt,:

By interpolation

Mirror scan velocity p

rofile

Rol

\j pitch, ya

wEp

heme

ris

(altitude

and

velocity)

Skew

Aspect ratio

by ne

ares

t neigh­

bor

Scal

ing

Mirror scan v

eloc

ity

prof

ile

Roll,

pitch, ya

w

Skew

Aspect ra

tio

Scaling

- re

move

sy

nthe

tic

pixels

External - projective ge

omet

ry

distortions

Map

Projections:

Spac

e ob

liqu

e Me

rcat

or (Opt,)

Map

Proj

ecti

ons:

UTM

Lat/

Long

State

plane

Map

Projections:

UTM

Spac

e ob

liqu

e Me

rcat

or

Albe

rs equal

area

Polar

stcrcographic

State

plan

e Lambert

conformal

North

American p

olyconic

Flexible rubber sheet

stre

tch­

in

g (A

rray

pr

oces

sor

planned)

Image

rota

tion

(A

rray

processor

plan

ned)

Perspective

correction

REGI

STRATI

ONDe

velo

pmen

t of reg

istr

atio

n parameters

Mapp

ing

of g

round

control

poin

ts

to entire sc

ene

or subscene

Digital

temporal and ma

ppin

g ov

erla

yImage

to image

Image

to map

Cursor de

sign

ates

li

ne and

pixel

of se

lect

ed fe

atur

esInteractive

sele

ctio

n of c

on­

trol

points (I

nclu

des

dis­

played GCP's

from m

aps)

One

pixe

l dot

for

GCP

location

(blinking)

Interactive

sele

ctio

n of c

on­

trol

points from su

bsce

nes,

merged for

full sc

ene

cor­

rection

(Up

to 100

GCP

max.)

Leas

t squares

fit

to a ma

ximu

mof

16 GCP's

(Opt.)

Tran

slat

ion

only

with

stan

dard

soft

ware

Least

squares

fit

to a

maximum

of 9

9 GC

P's

Computed surface

fit

by nth

orde

r polynomial

Subs

cene

only (Opt.)

Full

sc

ene

(Opt.)

Full

sc

ene

Full

scene

Vari

able

scales (Opt.)

Variable scales

Vari

able

scales

TABLfi

1RF.NI-RAI. PU

RPOS

E 1M

ACK

PROC

ESS IN

KPa

ge J_ of _3

(Con

tinu

ed)

GEOMETRIC

CORRECTION 6

REGISTRATION

ESL

IDIM

SCOMTAL SE

RIES

9

ISI

SYST

EM 470, SAND

CORR

ECTI

ONS

NoIntern

al sensor dis

tort

ion

(Lan

dsat

MSS)

Skew

Aspect r

atio

to

79x79

meter IF

OSc

alin

g fo

r li

ne p

rinter to

sp

ecif

ied

scale

Mirror sc

an v

eloc

ity

prof

ile,

ro

ll,

pitc

h, ya

w, de

tect

or

sampling d

elay

(planned).

Mirror scan v

eloc

ity

prof

ile

Roll

, pi

tch,

yaw

Skew

External - pr

otec

tive

geometry

dist

ortions

NoIm

age

rotation

Map

proj

ecti

on (p

lann

ed)

Rubber sh

eet

stretching

Imag

e ro

tati

onRound

earth

Flexible r

ubber

sheet

REGISTRATION

Development

of reg

istr

atio

n pa

ramete

rs

Mappin

g of g

round

cont

rol

points

to e

ntire

scen

e or s

ubsc

ene

Digital

temp

oral

an

d ma

ppin

g overlay

Image

to im

age

Image

to m

ap

Inte

ract

ive

cursor s

election o

f co

ntro

l points

Interactive

cursor location o

f co

ntro

l points

Manu

al se

lect

ions

of p

oint

s by

coordinates

Curs

or selection

of G

CPs

Least

squa

res

fit

to p

oly­

nomials

(up to c

ubic

)Le

ast

squa

res

fit

per

512x

512

sub-image

Full

scene

Yes

Entire im

age

by m

osai

c of

sub-

imag

es

Variable sc

ales

Vari

able

scales

TABlJ:

1.C.F.NF.RAI, Pl

IKPO

SI-

IMACI

: PUOCIiSSINC. (SOFTWARE

SYSTEMS)

Pago

1 of

(Continued)

GEOMGTRIC

CORR

ECTI

ON 5 RE

GIST

RATI

ON

CORRECTIONS

Internal sensor d

istortions

(Lands

at MS

S)

External - projective geometry

distorti

ons

KANDtDATS

Skew

Scal

e by

nearest ne

ighb

orVe

rtic

al skew al

so available

-JPL

VIC

AR

'Mir

ror

scan

velocity pr

ofil

e*R

oll,

pitch, yaw

*Skcw

Aspect ratio

Scal

ing

- by interpolation

Flexible rubber sheet

byfunction

Rotation,

translation, skew,

scale

by m

atri

x co

effi

cien

ts

Space ob

liqu

e Me

rcat

or

Imag

e ro

tati

on P

roje

ctio

n to st

anda

rd map

Flex

ible

rubber sh

eet

stretch­

ing

LARSYS

"Ske

wAspect ra

tio

Scal

ing

Image ro

tati

on Rubber

shee

t stretching

REGI

STRATI

ONDe

velo

pmen

t of reg

istr

atio

n pa

rame

ters

Mapping

of g

round

control

points to entire scene

or

subscene

Digital

temporal an

d ma

ppin

g ov

erla

yImage

to image

Imag

e to

ma

p

Not

in COSMIC

Manual co

ntro

l points

Man

ual

or interactive

sele

ctio

n of c

ontrol points

Cross correlation

of a

reas

around selected control

poin

ts

Man

ual

selection

of control

points

Computed

bloc

k co

rrel

atio

ns

for

12-30

blocks per

scene

2x2

tran

sfor

mati

on ma

trix

Tria

ngul

ar facet

surf

ace

fit

'Biq

uadr

atic

fit

to G

CP's

Partial

scen

e, limited

by

computer si

zeFu

llor

partial

scen

e*Full

scene

By control

poin

tVa

riab

le scales

"Yes

TABL

E 1

GENERAL

PURPOSE

IMAG

E PROCESSING

Page

of

OTHE

R SY

STEM

FUNCTIONS

GE IMAGE

100

BENDIX MDA

SI2

,S S

YSTE

M 10

1

GEOM

ETRI

C INTERPOLATION

TECHNIQU

ESNearest

neig

hbor

(O

pt,)

Cubic

conv

olut

ion

(Opt

.)Bi

line

ar (O

pt.)

Lagr

angi

an (O

pt,)

Hard

ware

/sof

twar

e im

plem

enta

­ ti

on a

llows

any

4x4

tech

niqu

e to

be

used

Nearest

neig

hbor

Soft

ware

implementatipn

of

any

4x8

technique

incl

udin

g re

stor

atio

n, cu

bic

convolu­

tion,

bilinear Lagrangian

(Opt

.)

Nearest

neig

hbor

Bilinear

Cubi

c co

nvol

utio

n (p

lann

ed)

FILT

ERING

High

pas

s, lo

w pa

ss,

Lapl

acia

n(O

pt,)

Th

eme

filtering via

Go lay

surrounds

(Opt

,.)

Sin

x/x

filter s

oftware

for

window d

isplay

(Standard),

hard

ware

fo

r fu

llsc

reen

(O

pt.)

"Restoration"

(Bcndix pr

opri

e­

tary t

echn

ique

)Any

line

ar fi

lter

(Sin

x/x,

high

pass, lo

w pass,

Laplacian)

Convolution;

Gaus

sian

High p

ass

Low

pass

Wiener

Homo

morp

hic

Dire

ctio

nal

Fourier

transform

Hada

mard

transform

TRANSFORMS

Four

ier

Hada

mard

Othe

r

NoYe

sYe

s (h

ardw

are

and

software)

Yes

(hardware/software

Yes

Yes

(har

dwar

e an

d software)

Categorical

tran

sfor

mSlant

TABLE

1GENERAL

PURP

OSE

IMAGE

PROC

ESSI

NG, P

age

3 of

3

(Continued)

OTHE

R SY

STEM

FUNCTIONS

ESL

ID1M

SCO

MTAL

SERIES

9ISI

SYST

EM 4

70, «A

ND

GEOM

ETRI

C INTERPOLATION

TECHNIQUES

Nearest

neig

hbor

Bili

near

Cubi

c co

nvol

utio

n

Near

est

neig

hbor

Bi

line

arNe

ares

t ne

ighb

orBi

line

arCu

bic

conv

olut

ion

FILTERING

Convolution

- using

arbi

trar

y kernal

Wien

erHo

momo

rphi

c -

Dire

ctio

nal

Ten

different

rout

ines

for

filter g

ener

atio

n Fo

urie

r tr

ansf

orm

NoConvolution

~ us

ing

arbitrary

kernal

LaPlacian

Robert's gr

adie

nt

Theme

filt

erin

g by Grow/Shrink

and

generalized

temp

late

mat

ch

TRAN

SFOR

MSFourier

Hadamard

Other

Yes

No

Yes

No

Yes

Yes

Discreet c

osin

e Discreet L

inea

r Ba

sis

Slan

t

TABLE

1GE

NERA

L PU

RPOS

E IM

AGE

PROC

ESSI

NG (SOFTWARE

SYSTEMS)

Page

3 of

3

(Continued)

OTHER SY

STEM

FUNCTIONS

KANDIDATS

JPL VI

CAR

LARSYS

GEOMETRIC

INTERPOLATION

TECHNIQUES

Near

est

neighbor

Near

est

neig

hbor

Bi

line

ar*Cubic convolution

*Spline

*Sin

x/x

'Nearest ne

ighb

or*Cubic

*Spline

*Sin

x/

x

FILTER

ING

Them

e cl

eanu

p by

grow-shrink

Rectangular

box

filter

3x3

medi

an filtqr

Grad

ient

, Ge

nera

lize

dLaplacian

3x3

linear filter

Wien

erCo

nvol

utio

n - fu

ncti

on se

lect

­ ed

by p

arameter in

put

(eg.

low

pass,

high p

ass,

Lapl

acia

n, directional)

Fourier

tran

sfor

m

*Convolution

*0ptimum constrained

rest

orat

ion

*Laplacian

'Gra

dien

t*F

ouri

cr transform

TRAN

SFOR

MSFo

urie

r

Hadamard

Other

*Not

in COS

MIC

Yes

Yes

'Yes

Yes

'Yes

'Yes

Slant

Discrete c

osin

eDiscrete linear b

asis

*Various co

lor

space

tran

sfor

ms

TABLE

2MULTISPECTRAL

CLASSIFICATION

Page

1 of

2

MULT

ISPECT

RAL

CLASSIFICATION

GE IMAGE

100

BENDIX M

DAS

I2S SY

STEM

1Q

1

SUPE

RVIS

ED C

LASS

IFIE

RParallelepiped (s

ingl

e-ce

ll)

Table

look

up (m

ulti

-cel

l)

Nonparajmetric j

naxi

mum

like

li­

hood (O

pt.)

Ha

rdwa

re/s

oftw

are

implementa^

tion

Hard

ware

bul

k classifier,

in*

elud

ing

table

look

up an

d pa

rame

tric

max

imum

li

keli

hood

(O

pt,)

Bayesian maximum li

keliho

od

Gaussian or

aggregate

Gaus

sian

standard

Maxi

mum

likelihood

Parallelepiped

Table

look

up (h

ardw

are

and

software)

UNSU

PERVISED

CLASSIFIER

Spec

tral space

clus

teri

ng

Region

gro

wing

in

image

space

Semi

-sup

ervi

sed

software(Opt,)

Yes

(Opt

,)Yes

- software - ca

n be

done

on c

urso

r-de

sign

ated

polygon

MENS

URATION

Pixe

l count

of s

trat

aFu

ll screen

Suba

reas

by

them

atic

mask

I'ot

al scene

Cursor e

nclo

sed

area

Interactive

manual d

igit

izer

Mu

ltip

le scene

areas

Subsccnc d

efined b

y cursor o

r re

ctan

gula

r wi

ndow

INTE

R-BA

ND MAN

IPUL

ATIO

NIn

ter-band functions

Spec

tral an

alys

is

Spec

tral Ax

is Rotations

Rati

oing

, di

ffer

ence

-ove

r-su

m,

norm

aliz

atio

n (hardware)

Rati

oing

(software)

+ - x

T (hardware

and

soft

ware

)

Eigenvector

(software)

Eigenvector

Regr

essi

onFactor

Eige

nvec

tor

(har

dwar

e and

software)

Karh

unen

-Loe

ve (h

ardw

are/

so

ftwa

re)

General

Axis R

otat

ion

(hard­

ware)

Karh

unen

-Loe

ve

General

Axis Ro

tati

onKa

rhun

en-L

oeve

(h

ardw

are

and

soft

ware

) General

Axis R

otation

TABLE

2MU

LTIS

PECT

RAL

CLAS

SIFI

CATI

ON (SOFTWARE

SYSTEMS)

Page

1 of

2 (C

onti

nued

)

MULT

ISPE

CTRA

L CL

ASSI

FICA

TION

KANDIDATS

JPL VI

CAR

LARSYS

»'

SUPERVISED CLA

SSIF

IER

Bayesian ta

ble

lookup wi

th

spatial

neig

hbor

bac

kup

*Bay

esia

n*Parallelcpiped

tabl

e lookup

'Par

alle

lepi

ped

table

lookup

with

Ba

yesi

an se

cond

ary

clas

sifi

er*Stored

table

Bayesian

Maxi

mum

like

liho

od for

each

pixe

lSample (p

er fi

eld)

cl

assi

fier

*Multiimage

layered

classifier

UNSU

PERV

ISED

CLASSIFIER

Spec

tral

sp

ace

clus

teri

ng

Regi

on gr

owin

g in image

spac

e

Euclidian distance in

me

asur

emen

t space

Not

at pr

esen

tYe

s

Boun

dari

es by

Roberts

grad

ient

and

adap

tive

th

resh

oldi

ng

*Aut

omat

ic bou

ndar

y fi

ndin

g/sa

mple

cl

assi

fica

tion

MENSURAT

ION

Yes

r subareas by b

inary

mask

Yes

INTER-BAND MAN

IPUL

ATIO

NIn

ter-

band

functions

Spectral analysis

Spectral axis rotations

'Not in

CO

SMIC

Algebraic

operations

+ - x

T av

erag

ing

Any

Fort

ran-

writ

able

fu

ncti

onRatioing

Eigenvector, by

tra

inin

g ar

eas

or on

al

l data

^Eigenvector

*Eig

enve

ctor

Karhunen-Loeve.

Also

fa

st K-L

General

Axis

Ro

tati

ons

*Karhunen-Loeve

General

Axis Rotations

*Karhunen-Loeve

TABLE

2MULTISPECTRAL

CLAS

SIFI

CATI

ONPa

Seof __2

(Continued)

MULT

ISPE

CTRA

L CL

ASSI

FICA

TION

ESL

IDIM

SCO

MTAL

SERIES

9I S

I' SYS

TEM

478*

, SA

ND

SUPE

RVIS

ED C

LASSIFIER

Adva

nced

stored ta

ble

(Bayesian)

Maximum

likelihood (a

rray

proc

esso

r)

Minimum

dist

ance

(Euclidian)

Para

llel

pipc

dParallelpiped

Bayesian

Table

look-up

Non-

para

metr

ic(h

ardw

are/

soft

ware

im

plem

enta

­ tion -

SAND

)

UNSU

PERV

ISED

CLA

SSIF

IER

Spectral sp

ace

clustering

Region

gr

owin

g in imago

spac

e

ISOCLS (iterative)

clas

sifi

er

T. can

be d

one

on cursor-

desi

gnat

ed su

b-ar

ea

No

Iterative

spatial

clus

teri

ng

based

on homogeneous

regions,

edge detection

and

enhancement

and

thre

shol

ding

MENSURATION

Yes

- by

st

i'at

a bi

nary

mus

k,

curs

or-d

esig

nate

d ar

ea,

or

inte

ract

ive

manual di

giti

zer.

Stereo m

ensu

rati

on (opt,)

Yes

- hardware 1/

30 sec,

Total

scene

or arbitrary

sub-

ar

ea (optional

real-time

mensuration

hardware)

INTER-BAND MAN

IPUL

ATIO

NIn

ter-

band

fu

ncti

ons

Spec

tral

an

alys

is

Spec

tral

axis ro

tati

ons

Sum, Difference,

Prod

uct,

Ra

tio

Sum, Difference,

Prod

uct,

Ra

tio

and

comb

ined

fu

ncti

ons

Ratio

Array

and

scal

ar a

rithmetic

Logical

operations

Algebraic

func

tion

s

Eige

nvec

tor

Eige

nvec

tor

Eige

nvec

tor

Karh

unen

-Loe

ve

General

axis ro

tati

onKa

rhun

en-L

oeve

General

axis

ro

tati

on

TABLE

2MU

LTIS

PECT

RAL

ANAL

YSIS

Page

2 of

2

MULTISPECTRAL AN

ALYS

ISGE IM

AGE

100

BF.NDIX MD

ASSY

STEM

*01

STAT

ISTI

CAL

OPERATIONS

Training f

ield selection

Deri

vati

on o

f training f

ield

statistics

Maximum number o

f sp

ectr

al

bands

anal

yzed

si

mult

aneo

usly

Maximum number o

f cl

asse

s

Optimum band s

elec

tion

Clas

sifi

cati

on s

peed

(p

er

512x51

2 pixel

4-band sc

ene)

Single p

ixel (c

ross

-hai

r),

rectangle, 45°

rotated

rect

angl

ePolygon

cursor (O

pt.)

Any

shap

e qu

adri

late

ral

as

small

as

1 pixel

with a

ny

orie

ntat

ion

Full scene

Any

shape

desi

gnat

edIn

tera

ctiv

e selection

Cova

rian

ce m

atri

ces

Hist

ogra

msHardware/software

impl

emen

ta­

tion

Time in

depe

nden

t of s

ize

Covariance m

atri

ces

Histograms (O

pt,)

Covariance m

atri

ces

4; 5 wi

th r

atio

ing

Up t

o 16

(O

pt.)

1614 - Di

spla

y processor

or 36 - Central

proc

esso

r

Any

numb

er,

8 themes d

isplayable

at o

ne t

ime superimposed o

n'

vide

o.

50,

disp

lay

hand

les

any number

simu

ltan

eous

ly64

Opti

onal

in

tera

ctiv

e feature

selection

software -

<4 o

ut o

f 16

max

.

All

input da

ta c

hannels

pro­

cessed u

nless

oper

ator o

ver­

ridden

Rati

os an

d di

verg

ence

analysis

Nomi

nall

y 3-

5 seconds per

class

(multicell signature)

CCT-

limi

ted

(I/O

li

mit)

20 m

in.

tota

l \

scen

e

10 c

lass

-

2 min. wi

th a

rray

processor

(lin

ear

with n

umbe

r of

classes)

TABL

b 2

MUL1

ISPE

CTRA

L ANALYSIS

Page

2 of

2

(Continued)

MULTISPECTRAL

ANAL

YSIS

ESL

IDIM

SCO

MTAL

SERIES 9

,.ISI SYSTEM-470,

SAND

STATIS

TICA

L OPERATIONS

Trai

ning

field

sele

ctio

nInteractive

selection

of

polygonal

areas

with

cursor

(21

subcommands)

Random sampling

Any

shap

e quadrilateral

up to

8 active

Derivation of

tra

inin

g fi

eld

statistics

Cova

rian

ce m

atri

ces

Stat

isti

cs file editor

Covariance m

atrices

Maxi

mum nu

mber

of

spec

tral

ban

ds

anal

yzed

si

mult

aneo

usly

Maxi

mum nu

mber

of cl

asse

s

Opti

mum ba

nd se

lect

ion

Clas

sifi

cati

on s

peed

(p

er

512x

512 pi

xel

4-ba

nd scene)

255

256

in one

pass

Yes

Manu

al

10 c

lass

-

52 se

c. wi

th A

SAP

(lin

ear with num

ber

of c

lasses)

Interactive

sele

ctio

n via

curs

or o

r ma

nual

sp

ecif

ica­

tion by

poly

gon

vert

ices

Designated b

y arbitrary

irre

gula

r boundary

Hist

ogra

ms

Cova

rian

ce m

atri

ces

(Har

dwar

e/so

ftwa

re implementa­

tion -

SAND)

14 256 wi

th speed

pena

lty

255

in one

pass

(8 pe

r pass at m

aximum sp

eed)

Non-

para

metr

ic

Bhattacharyya

470

- Ba

yes,

up

to 16

cl

asse

s,

disc

I/

O li

mit,

<20

sees

. SAND -

from m

emory, <2.5 sees.

TABL

E 2

MULT

ISPE

CTRA

L AN

ALYS

IS(SOFTWARE

SYST

EMS)

Pa&e

2

of

2 (Continued)

MULTISPECTRAL ANALYSIS

KANDIDATS

JPL

VICA

Ri'

LARSYS

STATISTICAL OP

ERAT

IONS

Trai

ning

field

selection

Manual on

di

spla

y re

ctan

gula

r area

Any

shap

e de

sign

ated

*P

olyg

on v

ertices

(may b

e fr

om

maps)

Rect

angl

e by

li

ne an

d column

Deri

vati

on o

f training field

stat

isti

csSm

ooth

ed Ga

ussi

an

Mean

an

d co

vari

ance

mat

rix

by c

lass

*Covariance

matrices or

manual

para

llel

epip

ed limits

Cova

rian

ce matrices fr

om

training fi

eld

or c

lust

erin

g re

sult

s

Maxi

mum nu

mber

of

spectral

bands

analyzed si

mult

aneo

usly

Maxi

mum nu

mber

of

clas

ses

Opti

mum ba

nd selection

Clas

sifi

cati

on sp

eed

(per

51

2x51

2 pixel

4-band sc

ene)

(Very

mach

ine-

depe

nden

t.

Included a

s illustrative only.]

* Not

in C

OSMI

C

Depe

nds

on mem

ory

avai

labl

e an

d length of

line

1630

1850

60

Operator selection

Bhattacharyya co

effi

cien

ts*Yes,

including ra

tios

Yes

- transformed

dive

rgen

ce

(ave

rage

over class

pairs,

maximized

minimum

pair

wisc

)

10 c

lass

es - CPU

time 45

seco

nds

- pa

rall

elepip

ed only

Appr

ox.

1.5

minu

tes

- paral­

lelepiped

with 2

clas

s Ba

yesi

an re

solu

tion

- al

l pixels 12 m

inutes - parallel­

epiped wi

th 10

class Bayesian

resolution -

all

pixels

TABLE

3SYSTEM IN

PUTS

Page

1

of

1

INPUT

DEVICE

GE IMAGE

100

BENDIX MDA

SI?

S SY

STEM

161

TV CAMERA

or e

quiv

alen

t (ima

gery)

Yes,

silicon

vidi

con

Opti

onal

vid

icon

Yes, vi

dico

n

MICR

ODENSITOMETER

Drum s

cann

er

Other

Optr

onic

s (O

pt.)

Optronics

(Opt,)

Optr

onic

s

Perk

in-E

lmer

10

10A

can

be

used w

ith

system

DIGITA

L TAPES

NASA

Landsat

Skylab S

I92

(Opt

.)LARSYS (O

pt.)

INPE

Landsat (O

pt.)

Airc

raft

scanner

(Opt

.)CCRS Landsat

(Opt

.)JSC

Universal

(Opt

.)HD

DT t

o CCT

conversion for

aircraft scanners (

unde

rconstruction)

Land

sat

Bcnd

ix M

^S scanner

$Sk

ylab

S1

92

$IN

PE (Brazil)

MSDS

(24

channel

scanner) $

BRIM s

canner

$

$ (C

CT a

nd H

DDT)

Landsat

ANAL

OG TAP

ESFM tap

e to

CCT c

onverter

for

Daedalus a

nalog

scan

ner

(developed a

nd o

perating)

(Opt

,

Opti

onal

No

TABL

E 3

SYST

EM INPUTS

Page

1 of

1

(Continued)

INPU

T DEVICE

ESL

IDIM

SCOMTAL SKRIHS 9

ISI

SYST

EM 4

70,

SAND

TV CAM

ERA

or eq

uiva

lent

(i

mage

ry)

Not

at ESL

facility

(sys

tem

opti

on)

Spec

ial

system op

tion

Yes,

vi

dico

n

MICRODENSITOMETER

Drum sc

anne

r

Other

Opti

onal

- us

er c

hoice

No

Pcrk

in-E

lmer

1010A

No

Opti

onal

(h

igh

spee

d 8-bit

video

digitizer

is included

as st

anda

rd co

mpon

ent)

No

DIGI

TAL TA

PES

Land

sat

Skylab S

192

Airc

raft

scanner

PCM

Land

sat

VICAR

Landsat

JSC

Universal

Skylab S

192

Swin

g (ISI software s

td.)

ANAL

OG TAPES

FM a

t ES

L facility

NoNo

TABL

E 3

SYST

EM INPUTS (S

OFTW

ARE

SYST

EMS)

Page

1 of

1 (Continued)

INPU

T DEVICE

TV C

AMERA or e

quivalent

(ima

gery

)

MICRODENSITOMETER

Drum sc

anne

r

Othe

r

DIGITAL TAPE

S

ANAL

OG TAPES

* Not

in CO

SMIC

KANDIDATS

Land sat

Skyl

ab S192

LARSYS

JSC

Universal

No

JPL

VICA

R

Inte

rfac

e software fo

r;

CRT

scanner

*Dic

omcd

im

age

diss

ecto

r sc

anne

r

Inte

rfac

e software fo

r:

I'erkin El

mer

1010

* Land sa

t *Skylab

S192

*Bcndix M2S

*NAS

A 24 channel

VICA

R

No

,,

LARS

YS.

«

*Int

erfa

ce so

ftwa

re

*0ptronics

LARSYS

* Lan

d sa

t *A

ircr

aft

scanner

*Sky

lab

S192

*CCU

S *JSC u

niversal

*VICAR

*Yes

TABLE

SYSTEM OUTPUTS

Page

1 of

2

OUTPUT

GE IM

AGE

100

BUND

IX M

DAS

1,2S

SYSTEM \p

l

VOLA

TILE

DISPLAY

Graphics

Two-

dimensional

grap

hic

disp

lay

CRT di

spla

y - grey s

cale o

r color

imag

e

Anno

tati

on

Tektronix

Yes

Yes

Cova

rian

ce m

atri

x Cl

uste

r pl

otti

ng

Hist

ogra

ms

Intensity

prof

iles

St

atis

tica

l nu

meri

cal

outp

uts

Mensuration

output

Tables of:

Cate

gori

zati

on gr

oup

accu

racy

Means

separation in feature

spac

eAr

eas

and

cont

ents

Co

vari

ance

mat

rix

Tran

sfor

mati

on co

efficien

ts

Cluster

plots

(pla

nned

) Histogram

(Opt.)

Cate

gori

zati

on ac

cura

cy table

Status mo

nito

r Hi

stog

rams

(hardware

and

software)

Intensity

prof

iles

Cl

uste

r pl

ots

(pla

nned

)

512x512

(480

lines

visible)

Four

8-bit ch

anne

ls su

per-

impo

sabl

e with e

ight

1-

bit

themes

Bias

, gain,

inve

rt fo

r al

l CRT

guns

Moving w

indow

(Opt

.)

Time

-lap

se d

ispl

ay (O

pt,)

La

rge-

scre

en display (O

pt.)

Ad

diti

onal

cha

nnel

s (O

pt.)

Hi

gh r

esolution

display. 850X

870

(Opt

.)

320x240

"moving

window"

3 channels,

3 bits/c

hann

el

3 bi

ts/c

olor

512x512

"moving window-scroll"

colo

r

Yes,

in

various colors

Yes

Yes

TAHL.'.

4SYSTEM OUTPUTS

Page

1

of

2 (C

onti

nued

)

OUTPUT

ESL

IDIMS

COMr

Al,

SI-R

ItS

9(I

SI SYSTEM,470,

SAND

VOLATILE D

ISPL

AY

Grap

hics

Two-dimensional

grap

hic

disp

lay

CRT di

spla

y -

grey sc

ale or

co

lor

image

N)

OB

Annotation

Tektronix

(T)

Grey level

disp

lay

(D)

Yes

(Opt

iona

l)

Cluster

(T)

Hist

ogra

ms (D

) Intensity

profiles (D

) Strata b

ound

arie

s from e

xt,

digitizer

(T,

D)

Scat

terg

rams

(T

)

Histogram

Covariance m

atrix

512x512

3, 8-bit

imag

es4,

1-bit

graphics,

color

"Scr

olli

ng"

Alte

rnat

e - field

TV stereo

view

ing

with 3

-D men

sura

tion

soft

ware

(O

pt.)

Flexible Isometric

Display

(Opt

.)

512x

512

4 8-

bit

images

4 1-

bit

grap

hics

"Scrolling"

Yes

- Co

lor

Yes

Yes

Cova

rian

ce matrix

Cluster

plots

Scat

terg

rams

Most

ot

her

syst

em ta

bles

640x

480

7 8-

bit

channels

(512x512 4

8-bit

channels wi

th8

1-bit

overlays,

dynamic

translation, scaling)

Scrolling, su

b-in

uiyc

se

lect

ion

and

reve

rsal

(SAND)

Pseu

do 3-D

isometric

disp

lay

with

variable rotation,

reli

ef,

incl

inat

ion

and

5xmag. (depth cueing -

SAND

) Large

scre

en display

(Opt.)

"Mov

ie Mode" available

with

video

disc

st

ore

(470)

Yes

- li

mite

d to operator

ente

red

free-hand

over

lays

(v

ia v

ideo

mask) (fully

soft

ware

su

ppor

ted

- SAND)

TABL

E 4

SYST

EM O

UTPU

TS (SOFTWARE

SYSTEMS)

Page

1

of

2 (Continued)

OUTP

UT

VOLA

TILE

DI

SPLA

YGr

aphi

cs

Two-

dimensional

grap

hic

disp

lay

CRT

disp

lay

- grey s

cale o

r co

lor

imag

e

Anno

tati

on

* No

t in COSMIC

KANDIDATS

No Pseudo 3-D

Yes

No

JPL

VICA

R

No

*Covariance

matr

ix

Clus

ter

plotting

*512

x512

co

lor

Comtal

* 10

24x1

024

B/W

Comtal

*640x512 B/W

RAMTEK

Yes

,' LARSYS

.'

No No Yes

(at

LARS and

in cosmic

software)

No (at

term

inal

s via

land

li

ne)

No

M

IO

TABLE

4SY

STEM

OUT

PUTS

Page

2 of

2

OUTP

UTGK

I MACK

100

BUNDIX M

DAS

SYST

EM 4

01

HARD

COPY

OUTP

UTS

Line

pr

inte

r

High

qua

lity

film -

grey sc

ale,

color

Over

lays

(Raster)

Maps (Line

Plotter)

Phot

ogra

phs

of display

Goii

ld pr

inte

r/pl

otte

rdonId

prin

ter

(Opt.)

Yes

Dicorned (O

pt.)

Ha

lfto

ne by

dot

matr

ix on

Gould

printer

Optronics

Inte

rfac

e for

Optronics

available

Geom

etri

call

y co

rrec

ted

and

scaled tr

ansp

aren

cies

(O

pt.)

Di

come

d film re

cord

er (O

pt.)

Geometrically

corrected

and

scaled

Optronics

Geom

etri

call

y corrected

and

scaled - Ta

pe output fo

r Dicomed

D163 (p

lann

ed)

Xynctics plotter

(Opt,)

Calc

omp

Planned

Yes

Yes

Yes

TABL

E 4

SYST

I-M OU

TPUT

SPa

8c

2 of _L_1Continued)

OUTP

UT

HARD

COPY O

UTPUTS

Line pr

inte

r

High quality fi

lm -

gr

ey scale,

color

Over

lays

(R

aste

r)

Maps

(Line

plot

ter)

Photog

raphs

of display

ESL

IDIMS

Yes

Also pr

inte

r/pl

otte

r

Yes

- ESL dr

um recorder

Interface

(B/W)

soft

ware

to

Uico

med,

Optronics

Up to 8x10 tr

ansp

aren

cies

ge

omet

rica

lly

corrected

and

scaled

Off-line Xynetics

Opti

onal

COMT

AI,

SHR1

P.S

9

No Yes

No Opti

onal

ISI

SYST

EM 4

70,

SAND

Electrostatic

prin

ter/

plot

ter

Laser

reco

rder

(Planned)

Opti

onal

- user-specified

brand

Up t

o 8x

10 t

rans

pare

ncie

s

Prin

ter /plotter st

d.

Xyne

tics

opt

iona

l

Yes

TABL

H

4SY

STF.

M O

UTP

UTS

(S

OFT

WA

RE

SYST

EMS!

' .P

age

2 of

2 (C

onti

nued

)

OUTP

UT

HARD

COPY OUTPUTS

Line

pr

inte

r

High qu

alit

y film - gr

ey

scal

e, co

lor

Overlays

(Ras

ter)

Maps (Line

Plot

ter)

Phot

ogra

phs

of d

ispl

ay

* No

t in CO

SMIC

KANDIDATS

Grey sc

ale

by overprint

Inte

rfac

e so

ftwa

re for

Di cor

ned co

lor

via

tape

.JPL

VICAR

Yes

Interface

software fo

r;

*Dicomcd color

PDS

flatbed

- Model

1010

Optr

onic

s dr

um scanner

Up to 8x10 tr

ansp

aren

cies

(off-line)

Calcomp

Sepa

rate

hi

gh-q

uali

ty

Polaroid output

'' LARSYS

''

Yes

*Yes

*Yes

*Yes

TABLE

5OTHER

SYSTEM FE

ATUR

ESPage

1 of

1

CH IM

AGE

100

Rl-N

DIX MDAS

SYST

EM,

101

MICROPRO

GRAM

MABL

E PR

OCES

SOR

Planned:

Spec

tral

pr

epro

cess

ing

Clas

sifi

cati

on

Trai

ning

Spat

ial

vide

o filtering

Them

e filtering

Yes

(MCP -

the

Rend

ix ha

rdwa

re

proc

esso

r)Pl

anne

d ar

ray

processor

for;

Supv.

clas

sifi

er

Geom

etri

c resampling

PRES

ENT

COMP

UTER

POP

11/3

5, 11/45

POP

11/7

0 (Opt.)

32K wo

rd /DO

S/35

or

45

64K word/RSX11/35 or 45

Up to 1 Mw RSX/11/70

(1 word =

2 bytes)

POP

11/3

5 (11/45,

11/7

0)

32K/

DOS

Up to 128K core words

RSX11M (Opt.)

(1 word =

2 bytes)

HP 30

00128K by

tes

core min

SOFTWARE

Lang

uage

Modu

lar

Inte

rfac

e mo

de t

o an

alys

t

FORT

RAN

IV (9

0%)

PAL

(10%

)FO

RTRA

N IV (8

0%)

MACRO

(20?

i)

Yes,

Yes

Batc

h (Opt.)

Interactive

Menu

Batc

hInteractive

Menu

Suppor

ts m

ulti

ple

user

s?Ye

s, multiple c

ompl

ete

user

st

atio

ns

FORT

RAN

IV (applications)

SPL

(sys

tem)

Yes

Batc

h Interactive

Menu (Opt.)

Comm

and

Yes

SPECIALIZED

FEATURES

Fast

I/

O for displayed

data

st

orag

e on

ta

pe -

8 sec. fo

r 512x512, 4

chan

nel

Area

ta

ble

outp

ut

- Land us

e by

designated area

Digi

tal

tape im

ages

E-W r

escan-

ned

by T

opo. qu

ad.

Nearest

neig

hbor

sc

an t

o us

ersp

ecif

ied

resolution

Pr

opri

etar

y interpol

atio

n

TAU

Ll-

5O

T1II-

R SY

STH

M

IMiA

TUUH

SPa

ge

1 o

f 1

(Con

tinu

ed)

MICROPROGRAMMABLE

PROC

ESSO

R

PRES

ENT

COMPUTER

SOFTWARE

Language

Modu

lar

Interface

mode

to

anal

yst

Supports m

ulti

ple

users?

SPECIALIZED

FEAT

URES

ESL

IDIM

S

ASAP

: ESL's

prog

ramm

able

array

proc

esso

r 1)

Ma

ximu

m li

keli

hood

2) Fourier

3) Re

samp

ling

4)

Cl

uste

ring

5) Spectral ratios

HP 3000 Series II

12

8K-

bytes

core min.

HP 21Mx also re

quir

ed w

ith

. AS

AP as

a

controller

50 M

By d

isc

min.

recomm.

FORTRAN

IV

SPL

(HP-

3000

)

Yes

Batc

h Interactive

Menu

Co

mman

d

Multiple complete u

ser

stations

Inte

ract

ive

resource inven­

tory system

(Opt.)

Map da

ta from e

xter

nal

coor

­ di

nate

di

giti

zer

Imag

e catalog

5 retrieval

systems

(Opt

.)

COMT

AL SKR1T.S

9

No POP

11/3

5

MACR

O 11

Yes

Interactive

Command

'isi

SY

STEM

*470, SAND

No - System 470

(yes,

vector,

matr

ix and

neig

hbor

hood

op

erat

ions

-

SAND

)

POP

- 11

/35

32K

words

11/7

0 optional

(1 word =

2 bytes)

FORT

H hi

gh le

vel

(85%)

code (15%)

Yes

- User ex

tens

ible

Batc

h Interactive

Command

(Men

u -

SAND)

Multiple complete us

er

stat

ions

(Opt., SA

ND)

Tree

- structured directory

for

images and

data

2

- 80MB d

iscs

standard

TABL

E 5

OTl

ll-H

SY

STliM

H

:ATU

Kl:S

(S

OI-T

WA

KI:

SYST

LMS)

I'agu

1

of

1 (C

onti

nued

)

KAND

IDAT

SJP

L V

ICA

RLA

RSYS

MIC

ROPR

OGRA

MM

ABtE

PRO

CESS

ORIB

M

2735

-MSP

fo

r co

nvol

utio

nNo

PRES

ENT

COMPUTER

POP-15 at KU

Could

go o

n any mi

ni-c

ompu

ter

24K

core m

inim

um

IBM

360/44

128K

core

360/65

1000

K core

360/67 un

der

CP 67

500K min

imum

SOFTWARE

Language

Modu

lar

Interf

ace mode t

o an

alys

t

Fortran

Fortran

IV (70%)

Asse

mbly

(3

0%)

Yes

Yes

Batc

hIn

tera

ctiv

eMe

nu a

nd C

ommand S

tring

Batch

inte

ract

ive

Command

Fortran

IV (8

0%)

Syst

em/3

60 A

ssembler (20%)

Yes

Batc

hInteractive

Comm

and

SPEC

IALIZED

FEATURES

Interactive

image

editor

Image

Based

Information

Syst

em

(IBI

S)Available

for re

mote

terminal

oper

atio

n Well documented so

ftwa

re,

with

trai

ning

aid

s

Not

in C

OSMIC

DEFINITIONS

These definitions are generally arranged in the order in which they occur

in the tables. Definition of all the terms used is not possible within

the scope of this paper.

GENERAL PURPOSE IMAGE PROCESSING (Table 1): The entire gamut of possible

operations on a picture, including radiometric and geometric alterations,

various diagnostic procedures and displays, image analysis, and enhance­

ments such as contrast stretching and spatial filtering. This includes

all of the processes leading to the extraction of useful information from

an image such as mapping, radiometric information extraction, multispectral

analysis and classification, calibration and application of calibration

data (rectification), change detection, and image data base operations.

This does not imply that all systems have all of these functions, nor that

the list is necessarily complete.

Radiometric Correction - Application of calibration data or cosmetic cleanup

to remove visible radiometric artifacts.

* Landsat debanding - Landsat imagery generally suffers from a banding

effect due to unbalance in the various sensors. This may be mitigated

by application of calibration data or by line-by-line cosmetic cleanup.

* Photographic Correction - Framing cameras produce non-uniformity of

radiometric response over the face of the images. This shading may be

removed if calibration data is available.

* Other - Pixel noise may be removed by frame-to-frame averaging.

36

Atmospheric radiometric distortions require special procedures to

estimate and correct.

Digital Level Slicing - Conversion of the brightness or digital number (DN)

continuum to series of steps in each of which an assigned DN is substituted

for a range of original DN. Each step may later be assigned a unique color.

Contrast Stretch - An alteration of the DN scale, usually to convert an

input restricted DN range to an expanded output DN range. This may be

linear or non-linear, as desired. Usually done by table lookup for

computer efficiency.

Geometric Corrections - All sensors will produce images with some geometric

warping. A primary example is the Landsat MSS, which, being a line scanner,

requires 14 types of corrections.

* Internal - The principal MSS corrections required are those for inter-

sensor pixel displacements, non-linearity of the scan itself and

horizontal/vertical unequalities in the pixel spacing. Also, for

framing cameras, raster, film, and lens distortions must be corrected.

* External - Correction for external effects such as spacecraft attitude

and altitude, projection onto the round Earth, Earth rotation, and the

various warpings required to register images to maps or to other images.

Geometric Warping - The geometric warping process is broken into two parts:

determination of the warping function required over the face of the image,

and the interpolations of DN required to generate the matrix of output pixels

37

Registration - Determination of the geometric warping function.

* Registration Parameters - Determined either by a mathematical modeling

of the warping desired, or by locating a series of control points (CP)

for which the true locations (e.g., on the ground "Ground Control

Points, (GCP)" or in a reference scene or framework "Relative Control

Points, (RCP)") are known both externally and by line and pixel.

* Mapping CPs to Scene - Specific location of the source of data for

each output pixel is required. The relatively sparse group of CP

locations over the face of an image are used, generally by some sort

of interpolation or mathematical formulation, to generate the required

continuous function.

* Temporal and Mapping - For temporal overlay, CPs are often determined

by cross-correlation of areas in the new and reference scenes. For

mapping, image-type references are not normally available, and CP

selection must be manual.

Geometric Interpolation Techniques - Source location for an output pixel is

rarely precisely at an input pixel location, so that some form of inter­

polation using a group of input pixels surrounding the required source

location is required to estimate the DN of the scene at the source location

Filtering - Spatial filtering is used to restore the high spatial

frequencies which have been attenuated in the imaging process, to sharpen

visible edges, to determine areas of high or low DN variance, to soften

noise effects, to remove banding, or for numerous other functions. Spatial

38

filtering is generally done by convolution, but may be accomplished by

multiplication in spatial frequency space via a Fourier transform. Golay

filtering minimizes isolated pixel noise in homograms (thematic maps).

Transforms - Normal image coding records the brightness at each discrete

instantaneous field of view, IFOV, in the scene as the DN in each pixel.

However, the energy in the scene may be grouped in other ways: the

Fourier transform displays the two-dimensional energy content of the

various spatial frequencies present in the scene; the Hadamard transform

displays a related quantity based on square-wave decomposition of the

image. Other transforms are possible and sometimes used.

MULTISPECTRAL ANALYSIS (Table 2): That set of operations involving the

analysis of multispectral data utilizing the spectral information.

"Multispectral analysis" is often used synonomously with "multispectral

classification," although it properly covers all of the required operations

involving multispectral data. Multispectral classification is the recogni­

tion of areas of uniform ground cover through recognition of the spectral

signatures (the multispectral vector of DN for a given material).

Classification usually results in the substitution in each pixel of a

"name" or color representing the ground cover class in place of the

original data.

Supervised Classifier - The computer is instructed as to the identification

of certain known materials in the scene by the use of "training areas".

It is then given decision rules, and instructed to display all pixels

matching the series of materials by unique codes (or colors).

39.

Unsupervised Classifier - The computer first groups into a series of

classes all of the pixels in the scene by measuring the tendency of

pixels of a given material to cluster around a given location of multi-

spectral space. Following this, the analyst identifies each material.

* Region growing - Pixels of a given material tend to form a series of

groups of contiguous pixels (in image space). Thus, "All wheat pixels

in the scene occur in a group of wheat fields, each of which is

identifiable." Region growing is the computer process of gathering

and displaying the pixels in the identifiable groups, after which the

analyst may identify each.

Mensuration - the determination of areas in the image or designated sub­

section. This normally is done by simple pixel count, although for some

purposes intra-pixel mixtures must be estimated.

Inter-Band Manipulation - For multispectral data, each spectral image is

a "band." Operations between bands have been found to be useful for certain

analyses.

* Functions - Various algebraic functions such as multiplication or

division are non-linear, and produce data of a new kind. For example,

division of a scene by a reference "flat field" scene, point-by-point,

will tend to normalize against variation of sensor response over the

face of the scene.

* Spectral Analysis - Analysis of selected data points to determine the

optimum processes to be applied to the entire image. Examples are

eigenvector analysis to determine the optimum band combinations to

40

allow multispectral analysis, or the selection of optimum ratios to

be used for display.

* Spectral Axis Rotation - Linear combination of bands, pixel-by-pixel,

amounts to producing a new projection of the data onto a set of

rotated axes. Rotation to the optimum angles following an eigenvector

analysis (Karhunen-Loeve) is the most usual, but other rotation angles

may be used for other purposes.

Statistical Operations - The set of processes and decisions required to

carry out the multispectral analysis.

* Training Field Selection - Operator designation of areas of known

materials or ground cover. May be done interactively or by off-line

mapping.

* Training Statistics - Characterization of the group of pixels of the

training area, usually as mean and variance in multispectral space

(covariance matrix). Other measures such as texture could be used as

a derived multivariate analysis "band" for use with the multispectral.

* Number of Spectral Bands - Analysis time goes up as the number of bands

used increases - but so may the classification accuracy. A tradeoff

to be made by the analyst.

* Maximum Number of Classes - For some classifiers, the time for classifi­

cation goes up with the number of classes. Another tradeoff, since the

analyst may group classes together as desired.

41

* Optimum Band Selection - Some bands, including possible ratios between

bands, will give better class identification than will others. Some

software is available to aid the analyst in this selection.

* Speed - Schemes using full analysis can be extremely slow; table lookup

methods can accomplish much the same task at greatly increased speeds.

Decision as to "optimum methods" has not been made, and may not be

possible; differences between methods is often secondary to differences

caused by variance in the scene or the analyst choice of training.

INPUTS (Table 3): In addition to the computer compatible tape (CCT) input

anticipated, other input devices will be useful for other than Landsat

MSS data.

TV Camera - For scanning input images, a local TV camera may be useful.

This normally will be a vidicon or image dissector.

Microdensitometer - For higher accuracy, either in spatial or in brightness

quantizations, various forms of travelling aperture devices are used.

These commonly take the form of either drum scanners or flatbed scanners.

Digital Tapes - The tape format expected for most MSS work is the "NASA

Landsat." Others, many of which are similar, may be encountered. NASA

and USGS are currently trying to define a universal tape format family

which will encompass many of the present formats.

42

Analog Tapes - For some purposes, image, data may be recorded on analog

magnetic tapes. Special analog-digital conversion is required to utilize

: this data.

OUTPUTS (Table 4): These may be either volatile on a terminal display,

or hard copy via line printer, image film, maps, or tables.

» Volatile outputs&

1 * Graphics - A number of graphics displays are available - these are

normally used for alphanumeric conversations with the analyst during

the analysis process.

* Two-Dimensional Graphics - Some graphics terminals have the ability to

display line drawings, graphs, cluster plots, and other diagnostic aids

* CRT for Image Display - When grey-scale or color images need to be

displayed during the analysis, a special television-like terminal

display is required. These displays must be continually "refreshed",

requiring a special memory in the system to continually generate the

required data.

* Annotation - Ability to superimpose grids, alphanumeric data and other

information on a displayed image.«t

j9 t Hard Copy Outputs

* Line Printer - The normal computer printer, which may also be used to

generate a grey-scale image by character selection and overprinting.

For a printer having the normal 10 characters/inch and 6 lines/inch,

43

a 1/24,000-scale image will be produced if the pixels are interpolated

to be 60.96x101.60 meters IFOV.

* Film - For most analysis, the desired output product will be high

quality film, either in black/white (B/W) or color. The latter will

normally be generated by photographic combination of three B/W films.

* Overlays - Film output requiring the ability to convert some set of

data such as polygon vertices to rasters.

* Maps - Output implying the capability of driving line drawing devices.

* Display Photographs - Usually for quick look or reduced-quality

permanent copy, limited in quality by the display device.

OTHER FEATURES (Table 5)

Microprogrammable Processor - Some of the analysis steps are very time-

consuming, even with fast general purpose computers. A number of special

purpose devices are available to perform specified process very efficiently.

These are usually used as peripheral devices to the normal computer in the

system,

Present Computer - The hardware systems discussed each are based on a modern

minicomputer. Depending on the degree of interaction, number of terminals

serviced, displays, and other features, the size required may vary consider­

ably.

Software - Most applications programs are written in Fortran. Various

higher-level languages may be used for parts of the system programs, and

for some of the applications. The feature to look for here is the ability

44

I

of the user to write and incorporate new programs as he wishes, since

all functions cannot be defined at purchase time, and the state-of-the-

art is continually advancing.

* Interface Mode to Analyst:

1. Batch - The analyst generates a series of commands, either at an

interactive terminal or via card job entry. The entire series is

run by the computer as a single job without further interaction

with the analyst.

2. Interactive Command - The computer carries out each command as it is

entered by the analyst, (usually) displays the results, and waits

for the next command. In some cases, it may be possible to assemble

or call a short series of commands (a procedure) which will be

accomplished before receiving the next command.

3. Interactive Menu - As each command is entered by the analyst, the

computer replies with a display which gives the analyst the choice

of a number of options. These might be parameter entries, function

options such as linear, log, table, or other defined types of con­

trast stretch, or entries relating to any multiple choice decisions

the analyst must make. This mode is of particular value to the

analyst unfamiliar with the details of the function being called,

as it guides him through the required decisions.

45

I HATP mTc512M8551988 MOSS, Marshall E.

AUTHOR ~~~

Water Resources in the 21st