Color reproduction of CRT-displayed images as projected transparencies

Theses

Thesis/Dissertation Collections

8-1-1994

Color reproduction of CRT-displayed images as

projected transparencies

Audrey A. Lester

Follow this and additional works at:

http://scholarworks.rit.edu/theses

This Thesis is brought to you for free and open access by the Thesis/Dissertation Collections at RIT Scholar Works. It has been accepted for inclusion in Theses by an authorized administrator of RIT Scholar Works. For more information, please [email protected].

Recommended Citation

Audrey A. Lester

B.S. State University College at Brockport (1978)

A thesis submitted for partial fulfillment

of the requirements for the degree of

Master of Science

in

Color Science

in

the Center for Imaging Science

in the College of Imaging Arts and Sciences

of the Rochester Institute and Technology

August 1994

Audrey Lester

Signature of Author

CERTIFICATE OF AFPROVAL

MS.

DEGREE THESIS

The

MS.

Degree Thesis of Audrey A. Lester

has been examined and approved

by two members of the color science faculty

as satisfactory for the thesis requirement for the

Master of Science degree.

Dr. Mark D. Fairchild, Thesis Advisor

Dr. Roy S. Berns

Rochester Institute of Technology

Center for Imaging Science

Title of Thesis:

Color Reproduction of CRT Displayed Images as Projected Transparencies

I,

Audrey Lester

I

hereby grant permission to the Wallace Memorial

Library of

RI.T. ,

to reproduce my thesis in whole or in part. Any reproduction

will not be for commercial use or profit.

Date:

--J..J...tt;j~/....(,-q

-I-L/ _ _ I

7

j

Audrey

A. Lester

Submitted forpartial

fulfillment

oftherequirementsforthedegreeof

MasterofSciencein ColorScience intheCenterfor

Imaging

Science inthe Collegeof

Imaging

ArtsandSciences

oftheRochester Instituteof

Technology

ABSTRACT

This thesis evaluated the color appearance predictions of four digital color

transforms

(Hunt,

RLAB, CIELAB,

and von

Kries)

between CRT "original"

images,

viewed in a lighted _room, and projected _slides, viewed in the dark. Calibrated projection of these images required resolution of several complex

issues. Upon projection, the slide colors changed. A rapid color shift (on the

order of _{75 seconds)} was followed

by

a _slower,

steady

degradation which also

had to beminimized for accurateimage presentation.

Therefore,

a model ofthe film behavior was based upon dye absorptivities and color measurements of

slides as

they

were projected.

The psychophysical experiments included a comparison between

preference choicesand_{memory matching} to theCRT

"original."

Two CRTwhite

points were evaluated: D93 and D65. The preference choices were, in

fact,

distinct from the selected matches. RLAB produced

statistically

superior matches over

any

other model for both white points. Model performance was

image dependent.

Occasionally,

CIELAB or von Kries images were equivalent

to RLAB.

However,

CIELAB and von Kries predictions ranged

widely

in their

performance. Hunt's image predictions consistently gave the worst results.

Interestingly,

RLAB was also elected as the most acceptable choice: judged 'acceptable'

inmorethan2/3ofall_cases,witha maximum approval of89%.

contributors. First of _all, Mark Fairchild and

Roy

Berns implemented the film

recorder model. Mark also wrote the C programs to calculate the digital color

transforms. Lisa Reniffprovided suggestions and support for

many metrology

concerns. ColleenDesimoneoversawthe

purchasing

issues.

Additionally,

thisproject wassponsored

by

awide

diversity

of government

and industrialfunding:

NSF-NYS/IUCRC

NYSSTF-CAT CenterforElectronic

Imaging

Systems

Management

Graphics,

Inc.

Xerox Corporation

Realization of such asignificant personal achievement could nothave been

possible without the assistance of

family

_members, preschool teachers and

friends who pitched in when

unsuspecting

crises arose. Thanks to Ruth and

John

Klipp,

Joan and Nelson

Lester,

Betty

Sheridan,

Roberta

DiNoto,

Amy

Stappenbeck,

and Anne

Hoenig,

who made it possible for me to reenter the

technical arena.

Most

importantly,

special recognition

belongs

to

my husband

and children

fortheir continual_{encouragement,}patience andsupport.

ListofTables xiii

1. Introduction 1

1.1. Visual Perceptionand

Cross-Media

ColorReproduction 1

1.2. OrganizationofThesis 1

2. Phase I-- CalibratedProjection

of35mmslides 3

2.1. Background 3

2.1.1. Film

Stability

andDegradation 3

2.1.2. SpectralChangesUpon Projection 5

2.1.3. Film Recorder Model 6

2.1.3.1. Beer-BouguerLaw 8

2.1.3.2.

Estimating

DyeConcentrations 8

2.1.3.3. Dye Concentrations as PredictorsofDigitalCounts 11

2.2. The EffectofProjection 14

2.2.1. Equipment 15

2.2.2. Experiment 1:

Resolving

SpectralDifferences 16

2.2.2.1. Initial

Study

16

2.2.2.2. Cool Down

Study

17

2.2.2.3. Results 17

2.2.2.3.1. Regression Correction 17

2.2.2.3.2. Fluorescence 21

2.2.2.3.3.

Cool-down

vs. spectral 22

2.2.3. Experiment 2: Evaluation Of Color Shifts 27

2.2.3.1.

Stability

ofthe Color

Change

27

2.2.3.2. Color

Change

With Repeated5 MinuteProjection 27

2.2.3.3. Results 28

2.2.3.3.1.

Chromaticity

Change Over Time 28

2.2.3.3.2. Repeated Projection 28

2.3. Film RecorderCalibration 35

2.3.1. Experiment 3: Characteristic Vectors 35

2.3.2. Experiment4: Tristimulus

Matching

37

2.3.2.1. SlideMeasurement 37

2.3.2.2.

Modeling

Design 38

2.3.2.3. Results: EvaluationofModel

Accuracy

39

2.3.3. Experiment5: Image

Variability

39

2.3.3.1.

Gray Variability

40

2.3.3.2. Precision

Study

43

2.3.3.3.

Processing

Variability

44

2.3.3.4. Slide Degradation DuetoProjection 46

2.3.4. Film RecorderSettings 49

2.3.5. Recommendations 53

3. Phase II--CRTtoSlideAppearance Model

Testing

55

3.1. Background 55

3.1.1. Visual System 55

3.1.1.1. Color

Matching

Functions andVisualPrimaries 56

3.1.1.2. Rod Intrusion 57

3.1.1.3. Background Effects 58

3.1.1.3.1. Helson-Judd Effect 59

3.1.1.4. Surround Effects 59

3.1.1.5. Cognitive Effects 61

3.1.1.6. Chromatic Adaptation 62

3.1.1.7. Incomplete Adaptation 63

3.1.2. Color ReproductionCalculations 64

3.1.2.1. vonKries 64

3.1.2.2. CIELAB 65

3.1.2.3. RLAB 67

3.1.2.4. Hunt 69

3.1.3. Field TrialsofColorAppearanceTransforms 72

3.1.4. Experimental Techniques 74

3. 1.4.1. Direct Comparisonof

Viewing

Techniques 74

3.1.5. Statistical Evaluation

Techniques/Theory

76

3.2. Experimental 79

3.2.1.

Facility

Design and

Set-up

79

3.2.2. Image SelectionandPreparation 82

3.2.2.1. Gamut Issues 83

3.2.3.

Modeling

Parameters 84

3.2.3.1. CalculationofGraysforImageBackground 85

3.2.3.2. CRT White Points 86

3.2.4. Experimental Procedure 87

3.3. Results andDiscussion 88

3.3.1 Preferencevs.

Matching

Choices

89

3.3.2 EvaluationofModelMatches 89

3.3.2.1.

Repeatability

96

3.3.3.

Acceptability

96

4.1 Conclusion 100

References 102

Appendix 1: RegressionAnalysis for Inter-InstrumentSpectral Correction 107

Appendix 2: SpectraoftheLeicaandKodakCarouselProjectorSources 110

Appendix 3: Colorimetric Data forFilm Recorder Calibration 112

Appendix 4: Tone ReproductionTables forFilm Recorder Model 114

Appendix 5: CRT Calibration Model 121

Model

Accuracy

123

Appendix 6: PreCalc Files With Intermediate Variable ValuesforHunt '93 and

Fairchild '91 Color Appearance Models 125

Appendix 7: Observer Instructions 127

Appendix 8: Observer Profiles andRaw Psychophysical Results 129

D93 130

D65 138

Appendix 9: CalculationsofZ-ScoresandConfidence Intervals 146

Experiment 1: All

Images, Preference,

D93 WhitePoint 146

Experiment 2: All

Images,

Image

Matching,

D93WhitePoint 150

Experiment 3: All

Images,

Preference,

D65 WhitePoint 154

FIG. 3. Spectraltransmittance for single-colorEktachrome slides 19

FIG. 4. Thermochromicshiftfor thegreenslide 23

FIG. 5.

AL*, ACab*,

andAHab*between

hot

andcoolslides 25

FIG. 6.

(a*,b*),

L*,

andCab* changeswhile yellowslidecools 26

FIG. 7.

Chromaticity

coordinatesof3slideswhile projectedover time 29

FIG. 8. Repeatedprojectionofthe magenta slide 30

FIG. 9. Color shiftbetweenprojected androomtemperature slides 31

FIG. 10. Colorchange after20minutes of projection 32

FIG. 11. 'Projected' dye absorptivities 36

FIG. 12.

Gray

slide_variabilityas afunctionofexposure/measurementday....41

FIG. 13.

Gray

slide_{variabilility} as afunctionof

surrounding

colors 42

FIG. 14. Colorerror compared with positionontheroll offilm 44

FIG. 15.

Variability

ofcontrol strips 45

FIG. 16. Slide degradationduetoprojection 48

FIG. 17.

Density

ofgraysfromvarious CalSets(Line

Scans),

andLUTs 52

FIG. 18. Dimensions and_spacingoftheexperimental area 80

FIG. 19. Image dimensions

during

matching

and preferenceexperiments 81

FIG. 20. Imagesusedinpsychophysicalexperiments 82

FIG. 21. Experimental _viewingconditions 88

FIG. 22. Preference choices vs. matchestoCRT

'original'

90

FIG. 23. D93

matching

experiment: z-scores andconfidence intervals 91

FIG. 24. D65

matching

experiment: z-scores and confidenceintervals 92

FIG. 25. D93 preference experiment: z-scoresand confidence intervals 94

FIG. 26. D65preference experiment: z-scoresand confidenceintervals 95

FIG. 27. Spectraofthe Leicaand Kodakprojectorsources Ill

TableIII. Comparisonoftwofilmrecorder models 39

TableIV. Color degradation

due

toprojection(numberof_observers) 49

Table V. Influenceoffilmrecorder settings on

gray

exposure 53

TableVI. Average AEab*

caused

by

_exposure,

processing

or projection 54

TableVII. Measuredmodelparameters 85

Table VHI.

Acceptability

as afractionofimage

matching

97 Table IX.

Summary

ofimage

acceptability

ranges 98

Table X. EvaluationofCRT_model,bothwhite points 124

1.1. Visual Perception and

Cross-Media

Color Reproduction

Imagesreproduced

colorimetrically may

notappearthe same when viewed

across differentmedia. Visual colorperceptionis acomplex combination oflight

stimulation, signal _processing, and cognitive interpretation.

Thus,

perception

can notbe

simply described

by

the reflectance of an object. A quantification of

the brain's final interpretation of color must be determined in order to

successfully

predict the appearance of an image when it is presented under

different viewing conditions.

Gauging

this interpretation has proven tobe complex. Documented color

appearance phenomena illustrate the unusual behavior of our perception.

Several effects,

directly

related _{to this work,} will be reviewed. In _{the past,}

because of these complexities, color identification and color difference formulas

were devised after _{stringently controlling viewing} conditions

(Bartleson,

1981).

Today,

with the technological

availability

for novice users to create images

across media, a _{growing demand} to produce colors that appear the same on

output as those created on a

CRT,

for example, has sparked current research to

evaluate color appearance models that predict color matches across different

media.

1.2. Organization ofThesis

Without accurate _modeling of the film recorder and careful attention to all

experiments are dubious.

Many

details

of this work

logically

fall into two

distinct

categories: those

involving

creation of calibrated slide images and

psychophysical

testing

of color reproduction formulas.

Therefore,

this thesis

consists of two sections. Within the

first

section

many

calibration issues are discussed. This work has quantified the effect of heat from the projector light

source,

variability

of single-color-slide colors

due

to

processing

and the film

recorder,

modeling

error from the CRT and film _recorder, and degradation of

the colors that mightbe expected throughouta psychophysical experiment with

15 observers. Some other effects have been

empirically

identified as concerns,

such as changes intransmittance for patches made with the same digital counts but surrounded

by

different colors. Several settings for the film recorder were

alsoevaluated.

The section on psychophysics develops several topics: specific factors that

affect the visual system's interpretation of the input stimuli, experimental

techniques, previous

testing

of color appearance _models, color reproduction models (inthis_experiment) andevaluation of results.

Many

theories have been developed to

quantify

the visual perception of

color. This thesis used a

memory

matching

technique to evaluate a

forced-choice, paired-comparison psychophysical experiment

using

color reproduction formulae from four models: Hunt

93, RLAB, CIELAB,

and von Kries. These

2.1. Background

2.1.1.

Film

Stability

and

Degradation

Photographic film is comprised of an

extremely

complex system of

chemicals

containing

multiple functional groups.

Many

possible types of undesirable chemical interactions caused

by

a_variety offactors can occurin this

system. The filmusedinthis experimentwas Ektachrome 100 Plus

Professional,

a chromogenic

(dye-coupling)

photographic system in which dyes are formed

from colorless couplers

already

incorporated into the film. Two types of color

changes have been well-documented in chromogenic systems:

bleaching

and staining (Wilhelm and

Brower,

1993).

Bleaching

(or

fading)

results whenever the

density

ofdye isreduced incomparisontotheoriginal

image,

while

staining

refers to the formation of unwanted yellow product. An explanation for

bleaching

begins with an _{understanding} of what causes color in these compounds. At the heart of all chemical processes is the requirement for

electrons to reside in stable electronic states. These states account for the

presence of absorption bands as well as potential chemical reactivity. For

example, a molecule can absorb a quantum_only ifthe

energy

of the quantum is

the same as the _{energy difference} between the current electronic state and a

higher state ofthemolecule. Some compounds with a series of

alternating

single

and double bonds (called conjugated

bonds)

are known to have

energy

capable of

absorbing

light with wavelengths

ranging

from 400 to 700 nm, thus

causing

the appearanceof color.

Wheneverchanges inthe electronic states _occur, theabsorption peaks change

as well. For_example,stereochemical forms often_{have different}peak absorption

wavelengths since the cis- _{isomer is}

more constrained

(requiring

more

energy

to

vibrate). If the conjugated

double-bond

structure of the chromophore is

shortened, greater

energy

is needed to push the electrons into the next stable

level. This means that shorter wavelengths (instead of visible _ones) will be

absorbed

(Zollinger,

1991). The dyewillbecome colorless. In_{effect, the} amount

of absorptionatthe original wavelengthisreduced.

Many

acidic types ofreagentswill add to the double bonds and degrade the

sequence ofconjugatedbonds. Even

humidity

(or_water) plays a vital roleinthe

fading

rates of dyes throughhydrolysis or hydroxylation. Water is also known

to facilitate the generation of hydrogen peroxide (that has been shown to be

present

during

_{many photooxidations)}

(Zollinger,

1991). Illumination

may

also

influence the degradation. The _energy derived from the absorption of

ultraviolet radiation can produce the formation of excited states in the

bonds;

theseare thenableto react with atmospheric oxygentoreducethe double bond.

Oxidation of the dyes or other chemicals inthe photographic _emulsion,

may

form stains that are yellow or brown.

Quinones,

for example are yellow.

Notably,

naphthoquinones have been formed from the degradation of azo dyes

with peroxide

(Nassau,

1985). (Color couplers

frequently

contain an azo-type

linkage.)

Heatwill accelerate the rate ofreaction

thereby

_causing, over a certain

In_summary,

many

causes are knownto affectthe rate of film degradation:

chemical

stability

of the color photographic _materials, storage conditions (temperature and

humidity),

display

conditions

(intensity

of

illumination,

duration of _exposure, spectral distribution of the

illuminant,

ambient

environmental conditions

(temperature

and

humidity)

and

quality

of the

chemical

processing

(incomplete

washing

orremoval of photographic products

-including

silver,

halide

ions,

ordevelopers).

2.1.2. Spectral Changes

Upon Projection

Knowing

projection of slides causes

degradation,

initial measurements of

single-color slides (for characterization of the film _recorder) were made on a

spectrophotometer

(using

both regular and total geometries). The slides were then projected and measured with a spectroradiometer. The spectral curves

were compared to determine which

geometry

could be used to estimate the projected spectral transmittance.

However,

spectral incongruities between the

projected andunprojected slideswere nonuniform and varied for differentslide

colors. These differenceswere later showntobe caused

by

projection. Themost

likely

cause would be thermochromism due to

heat

from the projector

light

source. Other possible explanations for the spectral changes could be

photochromismorhydrochromism (lossof water).

A thermochromic shift differs from dye degradation and stain formation

because it is reversible. This suggests a mechanism in which a shift in

equilibrium (catalyzed

by

heat)

causes aminor change inthe _chromophore, thus

literature

(Zollinger,

1991)

and evaluated in standard reference materials (Grum

and

Fairchild, 1985);

however,

no published literature on thermochromism of

photographic film could be found. This phenomenon will be discussed in

greaterdetaillater(Lester and

Fairchild,

1994).

2.1.3.

Film Recorder Model

As stated

by

Berns

(1993),

processed film

may

be considered as a first

approximation, a transparent medium

consisting

of dyed gelatin. Given this

assumption, the

relationship

between digital input to the film recorder and

resulting film transmittance was predicted using:

1)

the Beer-Bouger Law

(which relates transmittance to concentration),

2)

a modification of the

tristimulus _matching algorithm described

by

Allen

(1984)

and implemented

by

Berns

(1993a)

to determinethe concentrations ofdye on the

film,

3)

a non-linear

stage (to model the film's tone-reproduction _curve), and

4)

an

empirically

derivedmatrixtoacountforinter-imageeffects. Thisis shownin Fig. 1.below.

First of _{all, the} image is

digitally

represented

by

three planes

(red,

_green,

and blue). The digital counts are modified _non-linearly

by

the film recorder.

Interimage effects (or cross-exposure within film layers to improve image

fidelity)

influence the actual dye formation. From dye concentrations and

absorptivities, the Beer-Bouguer Law is used to

spectrally

reconstruct the

transmittance of the image color. Tristimulus values are then calculated from

the transmittance and the spectra of the projector light source. These steps

define the _relationship between digital counts and output color (expressed as

3digitalplanes: red,green,blue

Red DigitalCounts

m

c 1_ y_l

(actual)

E GreenDigitalCounts

inter-image matrix

(rowsnormalized

andsumto₁₎

BlueDigitalCounts

c

l_ y_l

(predicted)

SpectralReconstruction fromconcentrations_using

the Beer-BouguerLaw

u

s

'1

Si

QJ u

fa

E to

Wavelength

SpectraofProjector Source

ReflectedoffScreen

x^

H

Wavelength

X=kZ(S(X)*Tw*T(X))

Y=kI (S_{X)*

T{X)

*J{X))

Z=kZ(S(X)*Tw*-z(X))

Consequently,

to _accurately reproduce colorsthroughexposure

by

thefilm

recorder, the

derived

model _{is inverted} so that tristimulus values are used as

input and the required digital counts are calculated. An iterative searching algorithm was developed

by

Mark D. Fairchildto accomplishthis task.

2.1.3.1. Beer-Bouguer Law

Given the assumptionthat processedfilm

may

beconsidered a transparent

medium

consisting

of dyed gelatin and that the dyes do not scatter light or

fluoresce,

the amount of dye in the film canbe described

by

a simplification of

the Beer-BouguerLaw:

^measured

=

T\gO

*e

<*PX,c

+

^rrPXfm

+

cyDX/y)

(1)

where

T^

_measured approximates

T^mternai

(the internal

transmittance),

T^q

is the transmittance of the substate (in this case

Dmin,

a slide made from the

maximum digital counts), c is _{the concentration,} and

D^

is the spectral

density

(or absorptivity

)

of_{the cyan,}magenta,and yellowdyes.

Inordertousethis strategy,it is

necessary

toknow the

absorptivity

of each

dye. While these curves are available from the manufacturer for cool

measurements, anothermethod was needed toderive

absorptivity

for'hot'

slides

(since it will be shown that projected slides have unique dye absorptivities).

This was accomplished _using principal component analysis from measurements

ofthree sets of

ramp

dataandwillbe discussedinsection2.3.1.

2.1.3.2.

Estimating

Dye Concentrations

parts:

determining

the initial _matrices,

converting

concentration to

transmittance,

calculating

tristimulus values, and

iteratively

refining

the

estimate ofconcentration untilatristimulusmatchisachieved.

Firstof_{all, the}

following

matrices are defined:

Tisthematrixofcolor

matching functions

(a3x31matrix):

T= ^400 ^410 ^400 ^410 *700 ^700 2₇₀₀

(2)

Eisthespectral power distributionofthelightsource (a31x31matrix):

E= "400 0 0 0 "410 0 0 0 ^700

(3)

Disa31x31 diagonalmatrix with

d^

=

-2.3026 *

T^measured:

~dm

0 0

0 D=

*400

0

d.

410

0 0 d

700.

(4)

<E> represents the dye absorptivities as a function of wavelength (a 31x3

matrix):

0=

*

(yellow)400 ^(magenta)

0

400 Y400xXcyan)

xAyeUow) xAmagenta) Jpicyan)

^410 Y410 Y410

xAyellow) ^(magenta) jAcyan) S,700 Y700 Y700

cis thenormalized concentration matrix (3x1):

^yellow..

"

magenta

(6)

/

is the absorption spectrum for the color of interest (a 31x1 matrix). K is

the absorptioncoefficient expressed as the_{-log(T^measured}

/

T^go)-A

400

410

K₇₀₀

(7)

Concentrations arefirstestimated

by

thefollowing:

c = (TEDO)-!TEDf

(8)

Fromtheestimated concentrations andthespectral dye

absorptivity

_curves,

the transmittance can be derived from the Beer-Bouguer equation. The

tristimulus values are calculated from the transmittance curves.

Next,

the

differencebetweenthe predicted and actualtristimulus values

(AX, AY,

and

AZ)

areusedtoobtainacorrection_matrix,usingthe

following

formula:

Ac =

(TEDO)"1

At

(9)

where,

At=

AX

AY

AZ

(10)

and, Ac =

Ac

Ac

Ac

yellow..

magenta

(11)

Acyellow

/

Acmagenta

, and

AcCyan

refer to an iterative correction in the

concentrations. Anew concentration vectoriscalculatedfromtheexpression:

Cnew

=C0ld + (Ac*

0.4)

(12)

The algorithm iterates until an

acceptably

small At is derived. A multiplicative factor of 0.4 was used to assure convergence. At that point, a reasonable

estimateforthe concentrationhas been found.

Upon

implementation,

spectral data was difficult to obtain with sufficient

precision

necessitating

colorimetric measurements _using a tristimulus filter colorimeter. In _{this case, the} D matrix was calculated

by

setting

T^

_measured to

0.4and thefirstestimate ofdye concentrations was givenby:

ccyan

= 1_X/Xn

Cmagenta= 1"Y/Yn

(13)

cyellow= *" Z/Zn

2.1.3.3. Dye Concentrationsas

Predictors

ofDigital

Counts

The next _step is to determine the

relationship

between the dye

concentration and the known digital counts of the input images. After

estimating dye concentrations

(using

the tristimulus

matching

algorithm described above), three tone reproduction curves were derived for

gray

colors

(those

having

equal digital counts). A cubic spline interpolation method was

Levels

FIG. 2. Tonereproductioncurves

relating

digital countstodye concentration.

If the

dyes

in color film were independent upon _exposure, the tone

reproduction curves would be sufficient to describe the

relationship between

digital count and

dye

concentration. However as

described

by

Berns (1993):

"...color film relies on the interactionbetween

dyes

to

improve

color

reproduction fidelity. This well-known

inter-image

effect must be taken into account. This was accomplished

by

exposing

film with a factorial designwhere _{red, green,} and blue ramps were

individually

variedwhile theothertwochannelswereset at amedium

Interimage effects canbe seeninthedata throughexamination ofpredicted

dye concentrationwiththeincrease in digital countsofoneinterimage _ramp at a

time. If the dyes were

independent,

the concentration of the other dyes would

remain constant.

However,

as the exposure of animage becomes more _red, for

example,not

only

doesthe cyan concentration

dramatically drop

_off,butalso the

yellow and magenta concentrations increaseslightly.

If each of these changes were

linear,

a 3x3 transformation _consisting of

slopes would model the interaction. The diagonal elements would contain

relatively

large positive numbers while the off-diagonals would have small

negative values. The actual data are not so well behaved. In one _{study, this}

obstacle was overcome

by

replacing

the simple slopes of the hypothetical 3x3

matrix with a matrix of spline functions

(Berns,

1993). However in this _thesis,

error in the data points from _processing and measurement variations produced

erratic spline functions.

Modeling

error was

particularly

obvious for the

gray

colors. In_{the end,}a3x3matrix gavethebestresults.

The matrix was derived in the

following

manner.

Using

18 new slide

colors, actual dye concentrations (calculated

by

the tristimulus

matching

algorithm) and predicted dye concentrations (derived through the tone

reproduction _curves) were determined. As

previously

implied,

ifthere were no

inter-image effects, these two concentrations would be equal (within

experimental noise). A linear regression model was calculated for each dye

concentration. The model was constrained so that each row summed to one. This restriction was used to maintain

gray

balance. The

resulting

inter-image

"yellow...

''magenta

1.0542 -.0455

-.1332 1.0125 0.1207

0.0040 -.0081 1.0040

"yellow..

"magenta

cyan...

(14)

The adjusted concentrations were used to reconstruct the spectral

transmittance ofthe predicted slide color

using

the Beer-Bouguer Law.

Finally,

tristimulus values were calculated from the spectral

transmittance,

projector

source and color

matching

functions.

Applying

the inverse of _{this model, tristimulus} values are transformed to

transmittance then to dye concentrations. The inter-image matrix is inverted to

change_{spectrally derived}concentrations to themodified concentrationsused for

the tone reproduction

look-up

tables.

Finally,

an iterative _searching algorithm

relates this second set of concentrations to digital counts. These input values

should result in the correct exposure

by

the Solitaire

8xp

film recorder, on

Ektachrome 100 Plus Professionalfilm.

2.2. The Effect of Projection

Aswas stated_previously,projectionis knowntocause degradationof slide

film. Thus initial work attempted to find a measurement method which would

not degrade the

film,

yet couldbe correlated withthe spectral data of projected

colors. The colorimetric _accuracy ofthese images wouldbe critical for Phase II:

psychophysical tests.

First,

single-color slides were created on a Solitaire film

recorder, processed, and measured on a spectrophotometer

(using

both regular

and total geometries). These slides were then projected and measured with a

2.2.1. Equipment

A_varietyofinstruments wereused tomeasurethe film. Inorder to reduce

degradation from _projection, a BYK Gardner Color Sphere

(BKYCS)

spectrophotometer was used to measure the transmittance of the ambient

temperature

("cool")

slides. Thisdouble beam instrument is designedwitha d/8

geometry, a quartzhalogen

lamp

withanIR

filter,

a 10 nm

bandpass,

a concave

holographic

grating

spectrograph and a solid state silicon diode

array

detector.

Both regular and total transmittance can be measured on samples with a

diameter at leastthe width of35 mm slides. The slides were measured atroom

temperaturesincethe opticaldesignoftheBYKCSdoes not_{significantly heat}the

object

being

measured.

Later,

measurements were made while the slide was projected. A Photo

Research PR703A spectroradiometer

(PR703A)

was used to record the

transmittance of projected slides at 0 from the normal to the projection screen.

This is a single-beam instrument with a 5 nm

bandpass,

a holographic

diffraction

grating

anda256elementphotodiodearray.

An LMT-C1200 Colormeter

(LMT)

was used to measure tristimulus values

of the single-color, 2-color and 9-color projected slide test targets used for

modeling (this will be discussed further). The LMT was also used to

quickly

track _chromaticity changes in the projected colors. The projected light was

imaged

directly

onto the detector.

Finally,

the LMT was used to measure the

CRT channel chromaticities (needed for Phase II). "This instrument is a

tristimulus colorimeter with _nearly perfect filter correction. [It consists of four

tristimulus_value; onefor the blueregion and one for thered _region.] Itis more

accurate and precise than available spectroradiometers due to calibration and

signal-level difficulties."

(Fairchild,

1990,

AppendixC).

The projector _screen, a Clarion Wall Mounted model

rigidly

stretched on a

black aluminumframe with a vinyl matte white surface

(M1300),

was chosen to

providethemost

uniformly diffuse

reflectance over a

broad

_viewingangle.

A LeicaP-2200 projector witha

flat

field lenswas used. The temperatureof

a projected

gray

slide (witha visual

density

of

2.14)

was 52C measured with a

YSI Series 400 thermistor _using a Fluke 73 _multimeter,

following

the guidelines

from ANSI Z38.7.5-1948

(Woodlief,

1973). (This temperature is far below the

maximum recommendedtemperatureof

70C.)

2.2.2. Experiment 1:

Resolving

Spectral Differences

2.2.2.1. Initial

Study

Asetof 24 slides were createdwith a Solitaire-16 film recorder. Thecolors

were chosen to provide a somewhat uniform

sampling

throughout the CIELAB

color space. At

first,

the slides were unmounted and

directly

measured in a

BYKCS spectrophotometer _using both regular and total transmittance

geometries. Nextthe slides were placed in glass_mounts,projected onto a white

matte screen and measured with a PR703A spectroradiometer. This instrument

measures spectral radiance of the projected slide. Spectral transmittance values

ofthe filmwere calculated

by

dividing

the spectralradiance of each slide

by

the

spectral radiance of a blank slide (in the same orientation as the other

2.2.2.2. Cool Down

Study

A subset of the slides was chosen: dark _gray, light _{gray, cyan, magenta,}

yellow, red, green, andblue. These slides were

heated

inthe Leica projector for

5minutes thenmeasured inthe BYKCS (regular

transmittance)

as

they

cooled at

0, 15, 30, 45, 60, 120, 180, 240, 360,

and 480 seconds. Thespectral datawere used

to calculate CIELAB metrics (for CIE Illuminant A and the 1931 Standard

Colorimetric

Observer).

2.2.2.3.

Results

2.2.2.3.1. Regression Correction

Measurements under dissimilar conditions required resolution of two

separate issues: which _geometry, regular or total

transmittance,

would be

equivalent to the projected

transmittance;

and how differences between

measuring

devices could be resolved in order to compare the spectral data. At

first,

it was thought one spectrophotometric

geometry

would show a consistent

correlationwiththeprojected spectral transmittance measurements.

However,

it became

readily

apparent that the transmittance curves mismatched in a non

uniformfashionthatvaried for differentslide colors. Figure 3

demonstrates

this

dilemma for three colors. The projected transmittances were used as the

reference spectra for

AEab

calculations. Regular and total geometries were

measured with the BYKCS _using unmounted film at room temperature. As

shown in Table

I,

neither _geometry correlated well with the projected

transmittancecurves. Forthesixmost

highly

chromatic_{slides, the} average

AEab*

geometries. The _geometry that produced the lowest

AEab*

varied from slide-to

slide. Most

importantly,

the amount of mismatch was colordependent.

Table I. Measurement of spectral

differences

(in terms of AEa^*) between

BYKCSgeometries_usingthePR703Aspectra asthestandard.

*

indicatesthe

geometry

withthe lowervisualdifference

SlideColor RegularTransmittance Total Transmittance

Cyan *

4.32 4.58

Magenta * 4.02 5.72

Yellow 4.64 *4.34

Blue * 5.71 7.46

Green *

10.58 10.62

Red 5.13 *4.93

Overall Mean 5.73 6.28

Std Deviation 2.45 2.41

One potential cause for the spectral incongruities would be differences in

instrumental design (for example bandpass). A simple _way to simulate the

larger bandpass of the BYKCS when

measuring

with the

PR703A,

was to

appropriately average the spectral data from the PR703A measurements (at

every

2 nm), resulting in data points with a 10 nm bandpass for both

(al

PR703A

-Projected

-+-

BYKCS

-Regular

....o--

BYKCS

-Total

400

500

600

Wavelength

0.3-p a.

0.2-

/I

>^.

/

0.1-^S.&[

^^^_

^S#?

0^ "

1 ' ' ' ' ' ' I *-*

700

After _{this revision,} a regression method to correct for other systematic

spectral differences was examined

(Reniff,

1994;

Robertson, 1987;

and Berns

and

Petersen,

1988). This regression procedure uses a mechanistic model to

spectrally

represent the errors as a function of reflectance (or

transmittance)

factor. If a particular parameter is

statistically

_{significant, the corresponding}

measurement difference between the standard and test values can be

mathematically

corrected to improve correlationbetweenthe spectral data. The

full model includes the transmittance

(T)

(normalized

between 0 and

1), T^, T^,

T^, T^,

and the firstand second derivatives. Thesevariables have a _relationship

todifferenttypesofinstrumental errors: theintercept indicatesthe presence of a

photometriczero error (caused

by

stray

light,

calibration_error, ormaladjustment

of instrumental optics). Measured transmittance and polynomial terms of the

transmittance indicatephotometric scale error (caused

by

calibration error). The

first derivative of the transmittance indicates linear wavelength scale error

(caused

by

a displacement of the diffraction grating, _resulting in errors

proportionalto theslope ofthespectraltransmittance). Thesecond derivative of

the transmittance indicates a bandwidth error (caused

by

a variation in

constructionofthe slits _surroundingthe monochromator).

Inthe current experiment, a set of five filters: yellow, green,

blue,

neutral,

and didymium were measured (Hemmendinger Color

Laboratory

Filter Set:

HCL-68). The BYKCS data were considered thereference. The

filters

were then

backlitwith atungsten-halogen source and measured at

0

from normalwiththe

thermochromicspectral change. The raw PR703A dataweredivided

by

the data forthe sourcebefore theregressioncalculations weremade.

Regression analysis was evaluated

using

_stepwise,

forward,

andbackward

regression. Theregressionwas rerun

using all the variables suggested fromthe

forward

method. This time the partial student's t

-test was evaluated to check

significance ofthe variables. Theinterceptwas not_significant,so the regression

was repeated for the final time to give a R2 of 25%. A residual analysis of the model showed no trends and a normal distribution of the error. (Statistical

outputfromthefinalregression_{is included} inAppendix

1.)

A small correction for linear wavelength scale (the first

derivative),

bandwidth (the second

derivative)

and photometric error (T2 and

T5)

were

statistically required.

However,

theactual changein the spectral data produced an average

AEab*

of

only

0.32 (with a range of 0.08 to 0.46). The spectral

variations werenot explained

by

instrumental differences.

2.2.2.3.2. Fluorescence

Although _unlikely in slide

film,

another

possibility

for the spectral

difference couldhave beenfluorescence. Fluorescentdyes absorb visible

light

at

one wavelength and emit visible light at another wavelength. If

fluorescence

were present, differences in the light source

illuminating

the slides could produce inconsistentchanges inthe spectral curves. Asimple comparisonin the

spectral data when the slides were illuminated

by

monochromatic or

polychromatic lightwould indicate ifthiswere aconcern.

(Grum,

1980). Several

MatchScan II spectrophotometer with both monochromatic illumination and

polychromatic illumination. The spectral transmittance from both measurements were identical (within instrumental variability). There was no measurablefluorescence.

2.2.2.3.3. Cool-downvs.spectral

Another

hypothesis

for the color shift was the effect of temperature. The

slides were

noticeably

warm when removed from the projector (while this was not truewhen measurements inthe BYKCSwere made). A quick screen of the

medium-gray

slide was run

by

_{warming it} in the projector for 5 minutes then using the spectrophotometer to

quickly

measure the slide several times as it cooled. The spectral curves changed in the same regions where the discrepancies between the instruments were found. A closer examination of

cyan, magenta, yellow, red, green,

blue,

and two grays was designed with spectra measured at

0, 15, 30, 45, 60, 120, 180, 240, 360,

and 480 seconds after projection.

The color shiftforthe greenslide is illustrated in Fig. 4. Graph

(a)

depicts

the spectral differences when the slide was measured with the BYKCS at 0 and 480 seconds after

heating

in the projector. The second plot

(b),

compares the

slide while it was projected,

"hot",

and the "cool" slide (at 480 seconds). The thirdgraph

(c),

representsthe spectralchangesbetweenthe "hot"and

"cool"

data

for both

(a)

and (b). In this last

diagram,

the two lines were almost _parallel,

strongly suggesting the same phenomenon. (The temperature variation

0.3

(a)

BYKCS,

Hot

BYKCS,

Cool

400 500 600 700

Projected, Hot

BYKCS,

Cool

0.05-400 500 600

Wavelength

700'

FIG. 4. Thermochromic color shift for the green slide,

(a)

Hottestmeasurement

possible in the BYKCS compared with the same slide after

cooling

480 _seconds,

(b)

Measurement (with the

PR703A)

while the slide was _{projected compared} withthe same slide measuredwiththe BYKCS after

cooling

_{480 seconds,}

(c)

_The

difference

calculation (hot

data

-cool

data)

from both plots

(a)

and (b).

Notice

the

differences

exhibit the same pattern

strongly

The projected slide exhibited larger transmittance changes than the "hot"

data

measured in the BYKCS.

However,

the slides were

already

cooler at 0

seconds thanat a steady-state projected temperature.

CIELAB metrics were used to

document

the color changes of the eight

slides as

they

cooleddown. As showninFig.

5,

shifts inlightness

(AL*),

chroma

(ACab*),

and

hue difference

(AHa^*)

varied unpredictably.

Usually

the chroma

was greater for the cool slide. Sometimes thelightness decreased upon cooling.

Frequently,

the chroma andhue

differences

occurred inoppositedirections. The

angleofthechangeinAb*vs. Aa*plots variedforeachcolor. Inother_{words, the}

colorsdidnot shiftina singledirection (like _yellow)butvariedinhue angle and

chroma inan independentfashion. The change inlightness

(AL*)

and chroma

(ACab*)

became asymptotic, with a rapid change for the first 90 seconds which

started to level off at about 4 minutes. Figure 6 demonstrates these results for

theyellow slide.

From a calibration _viewpoint, the measurements from the

spectrophotometer (at room

temperature)

were undesirable. The actual colors

which would be viewed upon projection varied in an inconsistent fashion from

FIG. 5. CIELAB changesin lightness

(AL*),

chroma

(ACab*),

and hue difference

(AHab*)

as each slide cooled down. Noconsistenttrendisobserved.

Itshould

be

notedthatheat fromthe projector

may

notbe the

only

possible

explanation for the observed spectral shifts. Some other factor associated with

projection, such as photochromism or loss of water

(hydrochromism),

though

unlikely,

may

cause the change in the chromophore. For the purposes of this

work

however,

it is sufficientto documentthe spectralbehaviorofthefilmwhen

projected. Thus for lack of a better

term,

measurements

during

projection will

(a)

Ab*

-0.1-AL*

-0.25

AC*

0

100

200

300

400

Cooling

Time

(seconds)

100

200

300

400

Cooling

Time

(seconds)

500

FIG. 6. CIELAB changes in

(a*,b*)

_coordinates,

lightness

(L),

and chroma

(Cab*)

2.2.3. Experiment

2:

Evaluation Of Color Shifts

2.2.3.1.

Stability

ofthe

Color Change

Four slides were chosen to

determine

if the color of the slides would

eventually

stabilize upon projection. Dark and light

gray

were chosen for tone

reproduction, and yellow and green were usedsince

they

exhibited the largest AEab*

changes in the cool down study. The LMT colorimeter was used to

quickly

measure the

chromaticity

coordinates at 15 secondintervalsforthefirst3

minutes, thenmeasurements wererecorded at

4, 6, 8, 10, 15,

and20minutes.

2.2.3.2.

Color Change With Repeated

5

Minute Projection.

Anew setofslideswas preparedfor thisexperiment so that

they

would all

havethe same projectionhistory. This

time,

a subset of8 slides was used (dark

gray, medium gray, cyan, magenta, yellow, red, green, and blue). These more

saturated colors exhibitedthelargestcolor shifts.

First,

the slides were mounted and measured with the BYKCS (regular

transmittance).

Next,

each slide was projected for four _minutes, a

spectroradiometric _reading was taken with the PR703A and after a total of five

minutes in the Leica projector, the slide was removed. After

cooling

they

were again measured with the BYKCS. This process was repeated three more times

(to give a total projection of

10, 15,

and 20 minutes, respectively). The raw data

were divided

by

data for a blank slide mount before

calculating

CIELAB

2.2.3.3.

Results

2.2.3.3.1.

Chromaticity

Change OverTime

Since the color of the slides underwent a transformation upon projection,

the next question was whether this change would stabilize after some time

period. The LMT colormeter was chosen to

rapidly

monitor the _chromaticity

coordinates

(x,

_y) ofthe projector output. Figure 7 compares these shifts for the

yellow, greenanddark grayslides. Inall_{three cases,}a rapid color shift(of 75 to

90 seconds) was followed

by

a _slower,

steady

drift inchromaticity.

Apparently

because of _projection, two phenomena were

taking

place in which no

equilibriumpointwas established.

2.2.3.3.2. Repeated Projection

In Fig.

8,

the magenta slide results were enlarged to illustrate how the

CIELAB coordinates alternated between "hot"

and

"cool"

measurements after

5,

10,

15,

and 20 minutes of projection. Figure 9 compares this shift for all six

slides. Upon cooling, the colors almost return back to their original color.

Arrows indicate the general change ofAb*vs. Aa*over time. A drifttoward the

yellow (positive b*

direction)

implies formation of yellow stain caused

by

projection.

A multivariate Hotelling's T2 test determined that the population mean

betweenthe "cool" and

"hot"

(projected)

slides (for the

L*,

a*, andb*_coordinates)

were _{statistically} different in

every

case. This implies that the slides

had

0.478

0.476

0.474-(a)

Yellow

0.4698

hO.4694

0.469

0.4686

200

400

600

800

1000

0.348

0.346

(b)

Green

0

200

400

600

800

1000

1200

0.38-

0.376-

0.372-(c)

Dark

Gray

0

200

400

600

800

1000

1200

Seconds

0.386

FIG. 7. Simultaneous comparison of the x and _y

chromaticity

coordinates for

Ab* 8

*

Cool

? Projected

(Hot)

20 minutes

i?Na%.

10<S-C\a$^>^

5vT^^Oa^2

K 20

*15

*10

*5

-3 -2 -1 C) 1

Aa*

FIG. 8. Color shift (inCIELAB Aa* and

Ab*)

as the magenta slide is

repeatedly

heated up in the projector then allowed to cool. Both groups of measurements

show a shiftinthe +b*

direction

(yellowing

_staining) whichis observed forall other colors. (The original BYKCS

data,

before any

projection, is the standard

(0,0).)

The

(b*,

_a*) coordinates were also plotted for all eight colors

simultaneously

in Fig. 10. Plot

(a)

representsthe overall effect ofprojection. No

single direction of color shift occurs. The coordinates

demonstrate

a

transformation toward lower chroma, however sometimes

relatively large hue

6-1

2

-2- -6-10

4

Cyan

Projected

0

Projected 6-Ab* 2- -2- -6-10 V +

f

2

4

6

-4-2

0

6

6-

Red

2-t

-2-Projected -6: 10- ' i

Magenta

-4

0

t ' r

2--2: -6--10

6

2

-2 -6 -10

\

+ + Projected

Yellow

-4

0

A

a*

/

Green

Projected *

4-20246

a 0 Projected

6:

f

+ 2-+ -2--6:

Blue

10-i -i i

0

FIG. 9. Comparison of the color shift between hot

(projected)

and ambient temperature

(cool)

slides. The arrows

depict

the change over time

due

to film

100-(a)

During

Projection

50-1

t>* 0

-50-1

-100-A

/P

^

-50 -25

V

25 50

100

50

(b)

After Projection

(degradation)

*

Original

*

After 20

_min

b* o

-50

-100

-50 -25 25 50

FIG. 10. Comparison ofthe color change after 20 minutes of projection. Notice

that the colors of the slides after projection

(b)

have ALL become more yellow

(due to

degradation),

while the dominant shift for the projected slides

(a)

is

towardlowerchromaticity.

In comparison, Fig. 10

(b)

describes the change caused

by

degradation in

These slides have all become more yellow. The magnitude of the change is

smaller than that induced

by

projection and incontrast, all colors shifted

only

in the positiveb*direction.

Table II summarizes the

AEa]-,

after 5 and 20 minutes of projection. The

visual change due to thermochromism couldbe quite large (for example, green

was about 10

AEaj-,*

units) and at

times,

was reduced _slightly upon further

projection. The degradation _effect,

however,

appeared to be consistent over

time: with increased _{projection, the} colors shifted from the original to the

yellow, positiveb* direction. This experiment was repeated withslides in open

Table II. Spectral differences (in terms of AEab*) due to

staining

(cool)

vs.

thermochromism

(hot)

_using the original cool (room

temperature)

slide as the

standard. The differences were calculated from "uncorrected"

/

"regression

corrected" spectra.

SlideColor Amountof

projection

Originalto

Cool

OriginaltoHot

Glass Mounts

OriginaltoHot Clear Mounts

Red 5minutes 1.4 3.9

/

3.7 3.2

/

3.0

20minutes 2.4 3.4

/

3.2

Green 5minutes 0.36 10.6

/

10.3 9.1

/

9.0

20minutes 1.41 9.6

/

9.3

Blue 5 minutes 1.8 6.3

/

6.1 6.3

/

6.2

20minutes 3.8 8.7

/

8.5

Cyan 5minutes 1.1 5.1

/

4.9 5.8

/

5.6

20minutes 3.3 5.0

/

4.6

Magenta 5minutes 1.9 5.0

/

4.7 5.4

/

5.2

20minutes 4.1 7.2

/

7.0

Yellow 5minutes 0.78 4.5

/

4.1 5.3

/

5.0

20minutes 2.34 4.2

/

3.8

Dark

Gray

5 minutes 1.3 2.6

/

2.5 2.3

/

2.2

20 minutes 2.8 2.9

/

2.7

Light

Gray

5 minutes 1.3 0.41

/

0.33 0.45

/

0.39

2.3. Film Recorder

Calibration

2.3.1. Experiment 3:

Characteristic

Vectors

As discussed

before,

it is

necessary

to

know

the absorption properties of each dye in order to derive the tristimulus matches. While these curves are

available from the manufacturer for cool _{measurements,} another method was

needed to derive absorptivities for projected slides (since upon projection slides

change in color). This was accomplished _using principal component analysis (with _{SYSTAT software)} from measurements ofthree sets of

ramp

data. Single-color-slides created on the Solitaire-16 film recorder were chosen to produce

elevensamplesforeachseries

ranging

from fullcolor

(yellow,

_magenta,or _cyan) to white (no dye). The slides wereplaced in clear mounts and measuredon the BYKCS thenprojectedandmeasured with thePR703A. The data (400 to 700 nm in 10 nm

intervals)

were convertedto density.1 The firsteigenvectorfor eachset

of ramps was used to model the dye absorptivity. This vector explained a

minimum of 99.8% variance in all cases.

Consequently,

the model provided a

reasonable estimateforthedye

density

curves.

For comparison purposes, the predicted dye absorptivities for the hot and

cool vectors were normalized

by

peakwavelength. The eigenvectors are plotted in Fig. 11. The curveshapeofthe "hot"yellow and cyandyes are narrower than

the

"cool"

dyes. More

importantly,

the "hot" magenta dye

density

shifted

hypsochromically

10nm.

1

(a)Yellow

400 500 600 700

>-,

l-

0.5-

0-(b) Magenta

CO

__:

CD

Q

"O CD N

(Xi

E

400 500

(c) Cyan

600 700

400 500 600

Wavelength

700

FIG. 11. 'Projected' dye absorptivities. Normalized

(first)

eigenvectors

derived

The shape of these new _curves,

however,

seemed _{too pointed,} so a second

set of eigenvectors was generated from a

database

of 60 colors. Another

principal component analysis

using

an 'equimax' rotation was used to estimate

three eigenvectors (which corresponded to the three dye absorptivities). As

stated

by

Berns,

combining

the two sets of vectors through multiple linear

regression

(thereby

minimizing

sum of squares _error), produces the best basis

vectors when usedforprediction ofdyeconcentrations.

2.3.2. Experiment 4: Tristimulus

Matching

2.3.2.1.

Slide

Measurement

As

previously

stated, themodelfor thefilmrecorder requires exposure and

measurement of a factorial design of _{red, green,} and blue ramps. This was

accomplished

by

keeping

two digital counts at 96 and _varying the third color

from0 to255 insteps of32. Aset of9 grayswas also preparedinorder tomodel

the tone reproduction. Measurements of previous slides made from the same

digital counts were problematic. About 10% of the slides were too dark to get

spectral data with the PR703A. Inorder to overcome this

hurdle,

simultaneous

measurements of the slides with the PR703A and LMT Colormeter were made.

Slides which were too dark to integrate with the spectroradiometer, were

projectedforatleast75 seconds(thetime requiredforthe thermochromic shiftto

occur) before LMT tristimulus readings were recorded.

Using

the available

spectraldataandtheLMTtristimulusvalues,a

relationship

was

derived,

by

Roy

Berns,

between the calculated tristimulus values from the PR703A and the LMT

XPR703A

=0-184 ₊ (96.743 *

XLMT)

+

(2.325

*

YLMT)

YPR703A

=0-157+ (101.627*

YLMT)

-

(1.123

*

ZLMT)

(1)

ZPR703A

= (0781*

XLMT)

+(50.441 *

ZLMT)

2.3.2.2.

Modeling

Design

For the final _model, a simpler experimental design was used. Slides were

created with two colors surrounded

by

a

gray

having

a reflectance of

approximately

20%. Theseslides were shaped

similarly

to the images employed

in the paired-comparison psychophysical experiments of Phase II. A series of9

grays (from 0 to 255 in steps of

32)

was exposed

twice,

created on both sides of

the slides. Interimage slides were not prepared.

Instead,

two digital counts

were kept at 0 while the third varied from

64, 128, 192,

and 255 (producing:

dark _red, dark green, and dark

blue;

to full color). Measurements were made

with the LMT. In order to improve _variability in the

data,

the slide projector

was connectedtoavoltage regulator.

First,

the tristimulus values were corrected to estimate tristimulus values

from spectral data. Next a tristimulus

matching

algorithm was calculated in

order to predict concentrations2. The 2

observer was _used, with the projector

light source as the

illuminant,

and the transmittance of the Dmin was used to

estimate the transmittance ofthe substrate. While the substrate is

actually only

the gelatin, the Dminis areasonablefirstapproximation.

2 _{The tristimulus}

2.3.2.3.

Results: Evaluation

of

Model

Accuracy

The tristimulus

matching

algorithm was modeledwith twosets ofdata:

1)

only

the

gray

slides, and

2)

the nine grays as well as the three ramps of _cyan,

magenta, and yellow. Measured and predicted concentrations of a separate test

set of colorswere usedto calculateAEab*'s. Themore complicated_{matrix, using}

the interimage effects (while

maintaining

the

gray

balance),

produced better

results.

Table III. Comparison of two film recorder models. AEg]-,* of test slide colors using

only

a

gray

balancemodel or

including

interimageeffects.

For

ALL Color Slides

For

GRAY Slides

Average Range Average Range

Model from Grays 7.66 1.32- 17.44 1.74 0.42

-2.58

Model

Including

Interimage Effects 5.73 1.05

- 10.78 1.67

1.33-2.13

2.3.3. Experiment5:

Image

Variability

A wide _variety of factors affect the

ability

to

reliably

create an image.

Physical limitations of the film recorder,

processing

variability, and image

degradation can cause actual images to deviate from the predicted colors _using

2.3.3.1.

Gray Variability

As

previously

stated,

interimage

effects were evaluated

by

_usinga factorial

design,

of _{red, green,} and

blue

ramps. This was accomplished

by

keeping

two

digital counts at 96 and

varying

the third color from 0 to 255 in steps of 32. At

one pointin

modeling

thefilmrecorder,each

ramp

wascreated ona single slide.

In_all,a series of9slides

(red,

_green,

blue,

and2

gray

_ramps, aDminand3slides

containing miscellaneous colors for model

testing)

were created three times.

These slides were processed at the same

time,

buteach series was measured on

different days. The Dmin slide was used to normalize the tristimulus values

measured for all the other slides. On four _ramps, a patch with a

gray

section

(withequal digitalcounts of96 for_{the red, green,}and

blue)

wasexposed.

Examination of the coefficient of variation (the standard deviation divided

by

the mean), allows a comparison between normalized values of different

magnitude. Several combinations of the

gray

digital counts exhibited 5 to 10

times the _variability as the other colors: 0-0-0 (Cv =

11.8%),

64-64-64 (Cv =

17.4%),

and 96-96-96 (Cv = 10.0%). The

variability of the 96-96-96 slide was

examined in greater

detail,

since it was present on six different _slides, in the

same position. If the

daily

populations (6 _samples/day) of the normalized

tristimulus values are evaluated, some minor changes are observed (see

Fig

12).

However,

when the data are reevaluated

by

the color

ramp

of the slide

(red,

green,

blue,

or either gray), the variability for the

'color'

is

usually

muchsmaller

and more

importantly,

different from other colors (Fig. 13). The slide-to-slide

('color'

ramp) differences are dueto thephysicallimitationofthe CRTinthe film

neighboring

pixels. The

day-to-day

changes are a combination ofexposure and

measurement

differences.

This will be

discussed

in greater detail in the next

section.

Y_Variable = _NORMALIZED 'X' TSV

0.027 +

0.026 +

0.025 +

0.024 +

0.023 +

0.022

0.021 +

0.02 +

0.019 +

0.018 +

+ +

* j *

+ +

r__+__*

+ +

2

DAY

FIG. 12. Populations of 96-96-96

gray

patches separated

by

DAY of

Y_Variable = _NORMALIZED 'X'

TSV

0.027 +

0.026 +

0.025 +

0.024 +

0.023

0.022 +

0.021 +

0.02 +

0.019 + + +

0.018 +

* *

*_4._*

--+ + H + + +

--Blue Grayl Gray2 Green Group3 Red

SLIDE

FIG. 13. Populationsof96-96-96

gray

patches sorted

by

the

surrounding

COLOR

2.3.3.2. Precision

Study

The purpose of this

study

was to anticipate the amount of _variability that

couldbe expected when film was exposed and processed over time. A set of 8

colors was exposed four times on three different rolls of film over three days.

Each roll was also processed on adifferent day. The colors were chosen

by

their

anticipated positionin CIELAB space (based upon previous experience with the

LUT used). A color fromeach quadrant in

CIELAB,

as well as four grays were

used. For each rollof

film,

theeight colorswererandomized (and theorder was

differenteachtime).

Average tristimulus values of each slide (measured with the

BYKCS,

calculated for illuminant

A,

and the 2 _observer), were used to determine the

"reference"

color from which AE^'s were calculated. For all 95 _{samples, the}

mean AE^* was

only

0.46 with a range of 0.44 to 0.48 units per day. (One

outlier with a AEab* > 4 was removed, a _gray with a digital count of 150). The

colors which seemed most variable werethemiddle grays. A

gray

of150 digital

counts had a meanAEab*of0.99 (maximumof 2.03). A

gray

of 200 had a mean

of

0.65,

withamaximum of1.21.

Another issue was the effect of internal calibration of the film recorder.

These slides were randomized with a known sequence and calibrated

every

6

slides. Aplot ofthe maximum and minimumAE^* values over time (indicated

by

the slide number on the roll of

film)

was used to determine ifthe calibration

hadanoticable effect. Asshownin Fig.

14,

the colorchanges appeartobe cyclic.

2.3.3.3.

Processing Variability

The

variability

due to

processing

changes is great. In order to minimize

these effects,the same photofinisher (witha goodreputation for

quality

_control)

was always used. Upon special _{request, this} vendor (the RTT

Processing

Lab)

also supplied us with the process control strips used to monitor each batch of

film developed for this project. Adip-and-dunk

processing

style was used. The

film was placed on the same hanger as the control

strip

(to ensure a reasonable

estimateofthe actual chemicalbehavior ofthefilmwe received). Thedensities

2

1.8

1.6

* 1.4

W _1.2

i

5

0-8

O

0.6

0.4

0.2

0

DP

B

A?

_" |'

ii

H I I I I-m _in n o\ r

i.o_ot>a\rHcn_n--^a-\><

,-h,<,i,i,_{ICNCNCNCNC^JCO}

Slide Sequence

Minimum

a _Maximum

FIG. 14. Change in AEab* with position on a roll of film (recalibrated

every

6

of three steps:

3, 5,

and 7 on these control strips (chosen to reflect _visually

noticeable

densities)

were monitored _using the status A filter set on a MacBeth

model TD903 densitometer. Figure 15 provides a

history

of the change in steps

5. The dateofthecontrol

strip corresponding

to the slides usedfor

modeling

the

Solitaire film recorder and for creation of the samples used in the D65

psychophysicalexperiments are marked onthe plot. Alsomarked on _{the graph,}

arethecontrol strips oftheslides usedforthe precision study.

Precision

-3

to

4-Q

Film Recorder

Modeling

D65 Experiments

Date

The random fluctations in the

processing

(as exhibited

by

the

density

measurements) make accurate sample preparation

extremely

difficult. There is

no _way to control apriori t