Transient processing in chromatic induction

C’ision Re.

Vol. 33. No. 7. pp. 707-712.

1983

0042-6959:83,67070’-~~3.ou:0

Prin!cd in Great Britain

PergamonPrw Ltd

TRANSIENT

PROCESSING

SUSAN

IN CHROMATIC

INDUCTION

E. D~vtfs*, IREXE A. FMVRE and JOHN S. WERNER

University of Colorado. Laboratory of Psychology. Boulder. CO 80309, U.S.A.

Abstract-The non-additivity of steady backgrounds and superimposed test flashes was investigated with 25 observers using a hue cancellation procedure. The subjects task was to vary the ratio of two lights comprising an incremental test flash (AR/AG) so as to maintain a unique yellow hue (i.e. red-green qui~ibrium) in the presence of a lon~wave~ength adapting background. Measurements were obtained for four different luminances of the test flash (for each background), covering a range of t.E-2.0 log units. The adaptive hue shifts generally support Walraven’s tinding that the background affects the hue of the test flash only by way of altering the spectral sensitivities of the receptors, but it does not add to the hue signals from th; test-flash. .

Color

induction

Hue

Chromatic adaptation

non-additivity

Walraven (1976) proposed that adaptation to steady chromatic backgrounds results in both a setective attenuation of receptor sensitivities in accordance with the von Kries’ (1905) Coefficient Rule and the separate processing of incremental stimuli and steady backgrounds. The latter process implies a non-additivity of backgrounds and superimposed test stimuli such that the background alters the receptor spectral sensitivities (thereby shifting the hue of the test) but does not directly add to the hue signals from the test stimulus, Rather, the visual system uses the spatial and temporal transients that define the test stimulus to process it separately from the background. Walraven called this non-additivity discounting the background. In Walraven’s original study temporlti transients were provided by free scanning. but in his subsequent studies (Walraven, 1981; Werner and Walraven. 1982) the stimuli were flashed. Shorter test stimuli might maximize tem~ral-transient informatian, but somewhat longer test flashes have often been used to increase the refiability of hue judgements (Werner and Walraven, 1982). The paradigm from which Walraven derived his two-process model involved the presentation of a test stimulus (composed of a 66Onm. AR and a 540nm, AG, mixture) that was superimposed on a larger steady background (6~nm, R}. The observer’s task was to adjust the test ratio, AR/AG, to maintain a unique yellow. For each background, the luminance of the test stimulus was varied over a wide range. When only the test portion of the stimulus was analyzed, it was discovered

that the ratio

AR/AG

was

independent of the test luminance. Thus, there was a non-additivity of the test and background fields, Other data indicate that there are conditions in whieh *Author to whom reprint requests should be addressed.

Adaptation

Background

the b~i~kground is not comptetely discounted, i.e. there is an additive effect of the background (Larimer. 1981; SheveIL 1978, 1980, 1982). However, it has been shown with Walraven’s (1981) paradigm that nonadditivity holds over a wide range of adapting-background luminances (OS-S.0 log td) and background chromaticities (Werner and Walraven. 1982). Drum (19Xi) has recently proposed that since most of these experiments were conducted with a small number of observers, the differences between the results of Shevell and of W&awn may axially be based upon ~~d~vidua1di&rences between observers, Drum tested what might be the limiting case in Watraven’s theory: whether or not the hue of a dim test Rash could be canceiled by the additive effect of a background (i.e. wbether the appearance of an otherwise “green” test Rash could shift at low luminances of the test flash). He found that this shift in hue did occur for three of his five observers. His other two observers reported that “the hue of the test flash was always either greenish or so weak as to be uninterpretabie” (p. 960). Shevell (1978) aiso reported the former effect, although he noted that “often the subject found the AR = 0 case to be impossible” (p. f651f l-his was indeed true for Rve of the six backgrounds in his spatial and temporal transient condition. In general, Shevell (1978, 1982) does not find an additive effect for backgrounds beiow about 50 td, although Drum did with backgrounds as low as t0 td. Again, given the possibility of individual differences in the additive effect, these different results may not be surprising. Since Drum’s experiment was based on only five subjects, the range of individual differences could not be completely assessed. Moreover, Drum’s method (nomjnal scale, color naming) did not allow for a ciear determination of the magnitude of the deviations from complete discounting of the background. For

708

SCSAY E. DAVIES ZI trl

these re;Lsons. u’e have measured the non-additivity of backgrounds using Walraven’s (19s I) paradigin and a larger number

of observers.

The range of background approximately

the

luminances

same

range

used covered

as used

by

Drum

(19511.

.\IETHODS

Obsrrwrs

Twenty-five adults (aged 18-34) served as subjects. They had normal

trichromatic

Farnsworth-Munsell Dvorine

and

American

plates. All observers psychophysics

vision

according

to the

I00 hue test, as well as by the Optical

pseudoisochromatic

had limited

or no experience

and were naive as to the purpose

in

of the

experiment.

The stimulus

is schematically

Fig. I. It was a 0.6-1.8’ presented

A three-channel the stimuli

background optical

of channels

I:! nm

(Channel

half-bandpass)

2: Corning

screen. The spectral

I and 1, which were brought

by a beam-splitter

to form the test flash, were (Channel or

(2412;

the chromatic

1: Instruments

a broad-band

j-d = 640nm).

channel

provided

C2403:

id = 652 nm) and was brought

background

filter

The third (Corning

together

beam

The sessions began with IO min of dark adaptation. The observer’s task was to adjust a neutral density wedge (increasing or decreasing the luminance of the 540 nm component of the test flash) until the AR,‘AG ratio was at the red-green (R-G) equilibrium point. Measurements were obtained for six trials in a state of neutral adaptation (i.e. no background). This provided the only practice trials for the observers. The neutral indeed

condition to

that

with adaptation the

background

condid

Followthe observers viewed the

shift the red-green

equilibrium

point.

ing these measurements, steady background for 4 min. Red-green ratios were then measured in randomly presented blocks of 3 trials at each luminance level of the test tlash. There was a total of 9 trials for each luminance contrast. Contrast

was

defined

by

the

luminance

ratio:

(AR + AG)/R.

by a pellicle.

LOG

was compared

demonstrate

with

The radiance was controlled by neutral density wedges and filters that were placed in focussed and collimated beams. respectively. The spatial configuration of the test flash the test-flash

Procedure

ditions

system was used to present

shaped by a monochromator S-A;

(2 set on. 4 set off)

(R).

on a rear-projection

composition

by the inset in

(AR + AG) that was

as a flash superimposed

on a 9.8” uniform

together

shown

annulus

was controlled by a photographic field stop placed in a collimated beam (common to channels I and 2) that was conjugate with the projection screen. The head position of the observers was maintained with a chin rest (30 cm from the screen). Luminance calibrations were made at selected points with an S.E.I. photometer. The luminance at other points was measured relative to these points with a P-I-N 10 photodiode and photometric filter that was coupled to a linear readout system (United Detector Technology, Corp.).

AR

RESULTS

AND

DISCUSSIOS

For fifteen observers, the AR/AG ratios were measured at 4 different luminance contrasts, with a

(1, =640)

Fig. 1: Log AG is plotted as a function of log AR for each observer. The data for each observer are plotted relative to a line with a slope of unity. The origin of each function along the axis of abscissae is arbitrary.

709

Transient processing in chromatic induction

?I

5 0.5

LOG

AR

(X,=640)

Fig. 2. Log AG is plotted as a function of log AR for each observer. The data for each observer are plotted relative to a line with a slope of unity. The origin of each function along the axis of abscissae is arbitrary. The filled circles are the original data taken from Fig. I ; the open circles show the replication measurements.

background a pupil

of 2.63 cdim’

diameter

LeGrand.

(approx.

4.5 mm.

Drum

This

background

a larger

The

reported

an additive contrasts

0.3. 7.0, 3.7 and 17.6. Comparable backgrounds

in his Fig. AR AG

effect of the approx.

the

average,

from

the

neutral

background

the R-G the

are 0.5

background

condition

shifted

Fig. I. where log AG AR. According

is plotted

to Walraven’s

esis, the ratio

AR/AG

the

in psychophysical

about

the

whether

purposes

different

described. original in Fig. For

the

to

the

similar.

of

with

five

are presented

in

normalized

Measurements

luminance

the

were made

contrasts

eleven-point

(spanning

the second

(range

the

the background

luminance.

should

evident

be most

there is an additive to deviate

the high contrast

Thus,

we would from

the unit

points.

with

the deviations

point

using

a Wilcoxon

test (Siegel,

for

the

1956). The computations

not affected

by the manner

normalized.

With

not. however.

reject the null

non-additivity

fit the unit

subjects

If

proportion contrast than

A large apparent argued

additive

that

an additive’

ments under

the data

were

test we could (P < 0.05) of

I were obtained

possibility

observers reported

often

poorly

became

reported that

difficult.

(Shevell,

from

contrast

used were

unlikely.

equilibrium that

the

point

was

hue judgments

unlikely

that

reliable

for substan-

of the test Hash. This conclu-

by reports

from subjects in other

1978; Walraven.

of our lowest

of

adjust-

At these low contrasts

R-G

seems

1982). Most

be not

Many

in making

have been obtained

luminances

is not

it could

effect to have been clearly seemed

and

It

could

sion is corroborated Walraven,

contrasts

difficulty

the

defined

measurements lower

I. However,

these conditions.

they

tially

of the background

for an additive

This

contrast

for this test are

effect

the luminance

low enough manifest. our

of the background.

in Fig.

three subjects

Thus,

in these data.

in the data of Fig.

contrast

01 the slope estinvrtiorrs

The data presented

fall below

(SZ) they are above

of

studies Rekrhility

syscon-

effect

signed-ranks

hypothesis

show at low

these points

slope at all contrasts.

effect is not evident

to

Shevell

contrasts.

highest

in which

slope

the remaining

lowest

this distribution-free

two observers

the unit

the line. The data from

intervals.

the deviations

matched-pairs

from

from

to the pre-

slopes by more

slope for each observer’s

the data

deviations

are sufficient

(TM)

the low

We compared

that

tematic

the data are

functions.

observer

an additive

expect

points

the line and for another

to

at the lowest

four

trasts. For one observer

proportion

effect for a significant

point

complete

According

effect is a constant

Note

inten-

and that

the slopes of the individual

mean

was 0.96.

in the four-

It seems that

for

for the individual

0.55-1.1 I).

(1952). the additive

reliable

data are are quite

subjects

were

bars show 95% confidence

The mean slope calculated

of the observers.

point

indeed

was I.04

a

that were pre-

condition

for the individual

condition

estimate

AG vs

be 1.00. The data are plotted

by pinning

a unit

with

plotting

with a line of unit slope; the points

dicted line. Error 0.99

of log hypoth-

be invariant

log coordinates.

AR. the slope should each subject

as a function

point

observers

data and the replication 2. These two sets of data

slope calculated

the

chosen from the fifteen observers

used in the first condition.

viously

ex-

of

the data were reliable,

measurements

who were randomly

in

by a factor

non-additivity

should

sity. Thus, on double

from

naive

The mean slope for these same subjects

The results for the fifteen observers

points

we repeated

The shown

4.66.

was

and

range of 0.3-l 7.6) using the procedures

equilibria

change

who were unpracticed

periments

at eleven

contrasts

of I.2 and 4.6 cd,‘m’

(based upon

observers

research. To determine

6 mm).

were

luminance

I). On

ratio

chromatic

by

range used by

luminance

and 0.4, respectively

tabled

pupil.

is in the luminance

(1981) who

for Drum’s

value

75 td for comparison

since he assumed

background.

42 td based upon

the

1957. p. 106: approx.

with Drum

of

I98 I

importantly, point

is already

; Werner

the

and

luminance

below

that

in

SLSASE. DAVIESet al

LOG AR

(&=6401

Fig. 3. Log AC is plotted as a function of log AR for each observer. The data for each observer are plotted relative to a line with a slope of unity. The origin of each function along the axis of abscissae is arbitrary. The data in the top panel are based on luminance contrasts of 0.2, 0.6, 2.0 and 7.6; the data in the bottom panel are based on luminance contrasts of 0.1, 0.5. 1.7 and 6.3, Drum‘s

e.uperimcnt

where

3n

additive

effect

was

observers. Using ten additional naive observers, we did howcvcr repeat the initial procedure using lower contrasts. The contrasts used for five of these observers were approx. 0.2. 0.6, 2.0 and 7.6. The contrasts used for the remaining live observers were approx. 0.1, 0.5, 1.7 and 6.3. The results from these two lumin~tnce-contr~~st conditions are shown in separate panels of Fig. 3. The mean slopes for the data in the top and bottom panels were 0.9 I and 0.99, respectively. One may now see a tendency for the lowest point to deviate from the slope of unity. For 7 of the IO observers in Fig. 3 the reported for three of five

lowest contrast point deviates in the direction of an additive et%% Statistical tests were not conducted because of the small number of observers in each condition, however, it seems cfear that an additive effect is present at these very low luminance contrasts. There is a close fit for all other points to the line of unit slope.

Replications with other bac~roitfi~~ To determine whether these observations generalize to different luminances of the background. five observers were randomly selected from the first condition for replications with three additional backgrounds. These backgrounds covered a range of

LOG AR (Ids6401 Fig. 1. Log AG is plotted as a function of log AR for each observer. The data for each observer are plotted relative to a line with a slope of unity. The origin of each function along the axis of abscissae is arbitrary. Different symbols denote the data for the different background luminances.

Transient processing in chromatic induction -0.64 to 0.82 log cd: m’. which was approximately the same range that was used by Drum (1981; -0.51 to 1.26 log cdim’). The procedure was the same as in the previous conditions except that the luminance contrast of the lowest point was determined separately for each observer in order to obtain the lowest intensity for which reliable judgments could be made. R-G equilibrium measurements were also obtained at two higher luminances of the test flash. separated by approx. 0.6 log unit. Each data point was based on at least 6 settings. The results for the three additional backgrounds and the original background of 0.42 log cd/m’ are shown in Fig. 4. These data were normalized by vertical translations, pinning the second point to the predicted function of slope 1.0. The data clearly show the same pattern of results for all backgrounds. The mean slopes were 0.98, 1.07, I.01 and 0.98 for the -0.64. -0.24, 0.42 and 0.U logcdim’ conditions, respectively. Systematic deviations from a slope of unity are not apparent for any of the luminance contrasts and backgrounds.* CONCLLSlON

According to Walraven (1976, 1981). the relation between AC and AR can be expressed by the following equation: AG = AR g(R)

(1)

where y(R), the adaptation factor, only depends on the intensity (R) of the background. Shevell (1978,

*Although our backgrounds covered approximately the same luminance range that was used by Drum (1981). our methods were different. Drum focussed on whether an incremental stimulus having a dominant wavelength of 498 nm could shift in hue (i.e. whether AG could be cancelled by R) when presented at intensities from 0.6 to I.5 log units above increment threshold. To address this issue with our data, we calculated the intercepts of the linear equations that related AR to AC3 when these increments were expressed in terms of the percentage of the background luminance. The data used for these CJIculations were based on the three background conditions (with five subjects) for which the luminance of the increments was mdde as low as possible for each observer. (This condition is favorable for demonstrating an additive effect.) It would be predicted from the additivity hypothesis that for AR% = 0, there must be a value for AC”/, (to cancel the additive effect of the background) whatever the luminance level of the background. Non-additivity (discounting the background) would imply that when AR:; = 0. then AGo% = 0. since there is no additive effect to be cancelled. Thus, regardless of the background luminance, Walraven’s hypothesis (equation I) predicts an intercept at On/ This extrapolation is based upon the linearity of R-G equilibria (hue invariance of unique yellow) under moderate chromatic adaptation, an assumption that is justified by the results of Cicerone er nl. (1975). The average intercept obtained for the three conditions and five observers was -0.60/, with a range across observers from -3.2% to + 1.3%. Therefore. evidence for an additive effect is not apparent for these observers under these conditions.

711

1982). using a number of different conditions. gested that the more general formulation is: AG = [AR +f(R)] g(R)

sug(2)

wheref(R) represents the additive effect of the background. It follows from equation (1) that the slope of the function relating log AG to log AR should be 1.00, while equation (2) predicts a slope that is less than 1.00. Walrnven (1979) reanalyzed Shevell’s data and concluded that (except for points near increment threshold) the non-additivity hypothesis (equation I) was supported by Shevell’s own data from conditions in which spatial and temporal transients were available to separate the background from the test flash. (See, however, Shevell, 1980). Walraven (1979) suggested that these transients were clearly provided in Shevell’s (1978) spatial and temporal transient condition which involved a 60-90’ annular test flash that was presented on a steady 4’ background. More recently, Shevell (1982) has reported large additive effects. but his steady condition did not involve temporal transients. Our stimulus conditions tttdsimized spatial and temporal transients as in Walraven’s (1981) paradigm and in Shevell’s (1975) spatial and temporal transient condition. The resultant slopes were close to 1.00. the value predicted by Walraven’s non-additivity hypothesis. Thus, the data from our 25 observers lead us to conclude that with Walraven’s paradigm (except for the lowest contrast conditions) there is strong support for his non-additivity hypothesis. The existence of individual differences in non-additivity is supported by the deviations from the unit slope of some observer’s lowest contrast point. However, the possibility must also be considered that some of these individual differences are related to criterion shifts since most observers found the task difficult at these low luminance contrasts. It might also be noted that in comparing the data for the different observers it should be considered that different stimulus conditions could produce an entirely different spectrum of individual differences from that observed under our conditions. It is true that the visual system may not always be able to parse out the steady-state signal (Shevell, 1978, 1980, 1982) but it is remarkable that it does so over most of its range (Walraven, 1979, 1981; Werner and Walraven,

1982). Walraven

(1976) demonstrated

that

subtracts out the effect of quanta from the adapting field; these quanta have a minimal (additive effect) or nil (discounting the background) effect on the hue of spatio-temporally distinct test fields. The nature of this neural mechanism is likely to be of fundamental importance for understanding chromatic induction. there is a (presumptively)

neural mechanism that

Ackno,vledyenle,,ts-This research was supported by funds from the National Institute of Mental Health (Grant I R03 MH3409-01) and from the University of Colorado Institute of Cognitive Science. We thank Bruce Drum. Kenneth

711

Sushs

E.

Knoblauch. Steven K. Shevell and Jan Walraven for their comments.

REFERESCES

Cicerone C. M.. Krantz D. H. and Larimer J. I 1975) Gpponent-process additivity-III. Effect of moderate chtomatic adaptation. Visioti Res. 15, IIZj-I IX. Drum B. (1981) Additive effect of backgrounds in chromatic induction. Visiufl Res. 21. 959-961. Kries J. v. (1905 1970) Influence of adaptation on the effects produced by luminous stimuli. In ~aft~~uc~ der Ph,ssiologie des Menschen, Vol 3. pp. 109-182. Vieweg. Brunswick. Republished (1970). Sources of Co/or Science. pp. jOl-512. MIT Press. Cambridge. Massa chusetts. Larimer J. (I98 I) Red/green opponent colors equilibria measured on chromatic adapting fields: Evidence for gain changes and restoring forces. Vision Res. 21. 501-512. LeGrand Y. (1957) Liykf. Colotrr and Vision (Translated by

DAVIES

rr cd.

2nd Ed. Hunt R. W. G.. Walsh J. W. T. and Hunt F. R. W.). Chapman & Hall. London. Shevell S. K. (1978) The dual role of chromatic backgrounds in color perception. Vision Res. 18. 1649-1661. Shevell S. K. (1980) Unambiguous evidence for the additive effect in chromatic adaptation. Vision Res. LO. 637-639. Shevell S. K. (19SZ) Color perception under chromatic adaptation: Equilibrium yellow and long-wavelength adaptation. Visiotr Res. 22. 279-292. Siegel S. Nonparamenic Srarisrics for rhe Behaviorul Sciences. McGraw-Hill. New York. Walraven J. (1976) Discounting the background-The missing link in the explanation of chromatic induction. Vision Res. 16. 289-295. Walraven J. (1979) No additive effect of backgrounds in chromatic induction. Vision Res. 19. 1061-1063. Walraven J. (1981) Perceived colour under conditions of chromatic adaptation: Evidence for gain control by R mechanisms. V>sion Res. 21, 61 I-620. _ Werner J. S. and Walraven J. (I 982) Effect of chromatic adaptation on the achromatic locus: The role of contrast, luminance, and background color. t’isicla Ref. 22. 929-943.