- Email: [email protected]
surround components
Vision Res. Vol. 32, No. 6. pp. 1125-1130. 1992 Printed in Great Britain. All nghts reserved
Copynght
c
0042-6989/92 $5.00 + 0.00 1992 Pergamon Press Ltd
Visual Segmentation of a Heterochromatic Field Into Center/Surround Components* CHARLES
E. STERNHEIM,?
Received 4 December
RICHARD
1990; in revised form
P. PENN?
18 September
1991
The process of visual segmentation was studied by quantitatively estimating the apparent difference (i.e. the perceptual contrast) between two segments of a heterochromatic field. A chromatic test bar (Munsell 5R 6/6, 5BG 6/6, 5Y 6/6, or 5PB 6/6) was centered in an achromatic surround (Munsell N6) and the luminance contrast between the test bar and surround was varied between 0 and 39%. At equiluminance (0% luminance contrast) the chromatic border between center and surround was vague and perceptual contrast was minimal. When luminance contrast was increased slightly (5-15%) the border between center and surround was more distinct and perceptual contrast was almost two times greater than predicted by a model that combines spatial differences in chromaticity and luminance as orthogonal vectors. A luminance border may enhance perceptual contrast by increasing the salience of the center/surround hue difference. A vector model is consistent with perceptual contrast only at relatively high luminance contrasts, where hue differences play a relatively minor role in visual segmentation. Perceptual contrast
Luminance contrast
Figure-ground
INTRODUCTION Spatial differences in luminance and chromaticity combine to influence the visibility of an object in a scene. For example, the outline of a blue figure on a gray background depends upon the luminance difference between the two visual areas. KofIka (1935) observed that without luminance contrast there is a tendency for a blue figure to lose sharpness and with steady fixation to fade from view. Perception becomes stable only if the figure and ground are non-equiluminant. The change in visibility of a chromatic field as a function of luminance contrast raises the question of whether spatial differences in chromaticity and luminance are processed independently. The aim of the present study was to scale the combined effect of chromatic and luminance contrast and to test a vector model that assumes independent processing of chromatic and luminance differences. The vector model was developed by MacAdam (1949) to account for visual acuity measured with chromatic Landolt Cs against a neutral background. The same mode1 has been used by Kaiser, Herzberg and Boynton (1971) and Frome, Buck and Boynton (1981) to explain the distinctness of a border between two heterochromatic fields. In our experiment a chromatic test bar was positioned in the center of an achromatic surround *A preliminary report based on this study was presented at the 1987 OSA Meeting, Rochester, N.Y. tDepartment of Psychology, University of Maryland, College Park, MD 20742, U.S.A.
Segregation
and subjects scaled perceptual contrast. Chromatic contrast at equiluminance was converted into equivalent achromatic contrast (EAC) and combined with actual luminance contrast by orthogonal vector addition. Results show that in the range of luminance contrast between roughly 5 and 15% perceptual contrast is enhanced compared to the vector model. This suggests that spatial differences in chromaticity and luminance may interact to improve the segmentation of a field into center and surround components. The data are consistent with the results of other research that support the notion that chromatic contrast may be enhanced by luminance borders (Stromeyer, Eskew & Kronauer, 1989; Cole, Stromeyer & Kronauer, 1990).
METHOD Stimulus and apparatus
The test field consisted of a test bar located in the center of a neutral gray surround (Fig. 1, right side). Munsell (1976) papers were positioned so that each component of the test field could be illuminated independently. Munsell paper N6, containing a rectangularshaped aperture, functioned as the surround and was placed 56.5 in. from the subject. Munsell papers 5R 6/6, 5BG 6/6, 5Y 6/6, 5PB 6/6, or N6 were placed 6 in. directly in back of the aperture. Light reflected from the rear paper passing through the aperture constituted the test bar and traveled in the same direction as light reflected from the surround into the subject’s eye. When N6 was placed behind the aperture (N6/N6 condition)
1125
1126
CHARLES E. STERNHEIM and RICHARD P. PENN
An experimental session began with 8 min ot’ dark adaptation followed by 2 min of light adaptation to the test field. Adaptation was controlled almost entirely by the surround since it occupied the largest portion of the test field. After the adaptation period the subject used a minimally distinct border criterion to adjust the luminance of the test bar to match the luminance of the surround. The average of five independent settings was used to determine ~uiluminancc. In the trials that followed the experimenter varied Comparison luminance contrast in a mixed order. The subject was given no feedback about luminance contrast. On each trial the subject was instructed to scan acrass the test field for 3-5 set, avoiding steady fixation. The subject’s task was to rate the perceptual contrast contained in the test tieid by considering the degree to which the test bar appeared to separate from the surround. The FIGURE 1. The test &Id, on the right, consists of a test bar eaciased comparison bar viewed in darkness was used as a by a rectangular surround (not to scat+ The compatisonfield, on the standard and given a value of “IO”. The subject rated left, is a bar equal in size to the test bar in a dark surround. the test bar with any positive number, including any decimal number. Subjects were well practiced in making the test bar was unseen when the two grays were made numerical ratings before data were collected. Twelve equiluminant. A test bar was visible at equiiuminance levels of luminance contrast were presented 10 times each in every session. Viewing conditions were replicated only when the chroma~~ty of the rear paper differed in separate sessions 4 times in a mixed order. from the chromaticity of the surround.* A comparison In order to rate perceptual contrast, the subject bar (Munsell paper N6) of fixed huuinance was attended to the difference in appearance between the positioned to the left of the test field (Fig. 1) at the same area covered by the test bar and the area covered by the distance from the subject. surround. For example, a gray (N6) test bar with Munsell papers were illuminated by light originating luminance contrast in the range l-5% appeared lighter from a tungsten-halogen source flltered through a combination of four Kodak filters (82C, 82C, 82A, than the gray (N6) surround. However, the difference in 8C). The fWred light was 6060 K and the CfE 1931 lightness was very slight and the test bar and surround chromatic&y coordinates (x, y) corresponding to the practically merged with each other. Perceptual contrast Munseli papers were as follows: 5R 6/6 (0.420, 0.363); was typically rated under 0.5. In comparison, a red f5R SBG 6/6 (0.266, 0.375); 5Y 6/6 (0.431, 0.452); 5PB 6/(i) test bar with 22.7% luminance contrast looked considerably lighter and redder than the surround. Since (0.272, 0.313); and N6 (0.333, 0.364). The surround and comparison bar were fixed at 1.43 log cd/m*. The the test bar appeared to separate more clearly from the luminance of the test bar was varied from equiluminant surround, perceptual contrast was typically rated close to the surround and up by turning a circular neutral to 10.0. density wedge located at a focal plane of the test bar optical pathway. Twelve levels of kminance contrast between test bar and surround were produced: 0 (equihtminance), 1.4, 2.8, 4.6, 9.2, 13.7, 18.2, 22.7, 27.0, 31.2, 35.3, and 39.3%. [Contrast = (I,,, - &,J(L,, + I+,&_] Color temperature, chromaticity, and luminance were calibrated in situ with a Minolta Chroma Meter (~S-~~). Procedure The subject, with a chin rest to help stabilize head position, viewed the test field and the comparison bar monocularly without an artifkiai pupil or achr~at~ng lens, The outer edges of these fitds aided in accommodation. of Mumull papars to form the test &Id in this mamar is similar to the experimental set-up that was used by Hess and Pretori (1894, as described by Hurvich dr Jameson, 1966) to study
*The positioning
brightness contrast.
The two authors served as subjects. They each have normal color vision and normal acuity with correction. CES had previously participated as a subject in psychophysical experiments. Since this was the first psychophysical experiment for RPP, he received extensive training in prelimina~ work where the procedures of this experiment were refined, Both subjects had knowledge of the experimental questions, but they were kept uninformed about their results during an experimental session. REWLTS
The contrast rating functions for GRAY/GRAY (N6/N6) given in Fig. 2 are based on averages from four sessions. There were only minor deviations in the results of individual sessions. Luminance contrast below about 25% is perceived as a lightness difference between center and surround, whereas luminance contrast above
VISUAL SEGMENTATION
:
‘4 - ORAYIGRA~ 12- . S:RPP
s F
lo-
:
1127
FIELD
+
EGGRAY
+
YlGRAY
-
P&GRAY
8-
ii) d
6-
g
4-
8
OF A HETEROCHROMATIC
20
O!
0
10 LUMINANCE
20 CONTRAST
30
40
(K)
FIGURE 2. Contrast estimates in the GRAY/GRAY condition. Each data point is an average of 40 ratings. 95% confidence intervals, based on total variance (n = 40), larger than the data symbols am shown by
error bars. The broken lines connect the data points.
approx. 25% is perceived primarily as a differece in brightness.* This change in appearance from a surface
mode to an aperture mode occurs near where there is a break in the contrast rating function. The dual-branch functions that best fits the data of each observer are shown by solid lines. If P represents the contrast estimate and L represents the luminance contrast, the functions for CES’s data [Fig. 2(a)] are log P = 2.43 log L - 2.50 for the lower branch and log P = 0.91 log L - 0.38 for the upper branch. The functions for RPP [Fig. 2(b)] are log P = 2.99 log L - 3.42 for the lower branch and log P = 1.48 log L - 1.27 for the upper branch. The broken lines in this figure connect individual data points. The perceptual contrast of an equiluminant chromatic test bar varies markedly with test bar chromaticity (Fig. 3). R/GRAY has more contrast than BG/GRAY, followed in order by Y/GRAY, and PB/GRAY. The contrast rating functions increase with luminance contrast and the four functions converge at the higher levels, where the test bar is clearly brighter than the surround regardless of test bar chromaticity. The EAC of a chromatic test bar is the luminance contrast which would produce, in the GRAY/ GRAY (N6/N6) condition, the same rating that
*Luminance contrast between 1 and 5% is above threshold, but the contrast estimates in this range are too low to plot clearly above the horizontal axis.
0
10 LUMINANCE
20 CONTRAST
30
40
(%)
FIGURE 3. Contrast estimates for R/GRAY (triangles), BG/GRAY (squares), Y/GRAY (pluses), and PB/GRAY (diamonds). Each data point is an average of 40 ratings. The EAC values are as follows: R/GRAY (16.3 and 18.2% for CES and RPP, respectively);
BG/GRAY (13.7%, 12.8%), Y/GRAY (7.8%. 7.2%); PB/GRAY (2.4%, 3.0%).
the equiluminant chromatic test bar produced. If P represents the contrast estimate, EAC is obtained by using P =f(L) (L being the luminance contrast and f is the dual-branched function of Fig. 2). An EAC for each chromatic test bar is given in the legend of Fig. 3. The vector model predicts that P,, =f [(EAC' + /2)o.5]. For example, the average rating given by CES to an equiluminant R/GRAY test bar is 2.8 [Fig. 3(a) triangles, luminance contrast = 01. The EAC of R/GRAY for CES is 16.3%. If 9.2% luminance contrast is introduced for this test color, the total effective contrast [TEC = (EAC’ + 12)o.s]is 18.7% and P,, is 3.9. Empirical estimates of contrast assigned to R/GRAY conform with the model (Fig. 4), except for a limited range of luminance contrast, between roughly 5 and 15% where contrast estimates are comparatively high. Although the disagreement with the model is not very marked, enhancement of contrast at moderate levels of luminance contrast appears to be a reliable effect for both observers. Since EAC varies with test bar chromaticity, the shapes of the theoretical contrast rating functions, as plotted as in Fig. 4 with actual luminance contrast as the abscissa, are expected to vary. However, theoretical functions are expected to have the same shape if contrast estimates are plotted against total effective contrast
1128
CHARLES
E. STERNHEIM
and RICHARD 30
14
-I
P. PENN
S:CES
12 10 8 PEYGRAY
6
MODEL
4
-
MODEL
0 0
12
z
10
3
6
& d 5
6
8
2
30 40 CONTRAST (%)
50
FIGURE 5. Contrast estimates for R/GRAY (triangles. broken line, vertical offset 12 units), BG/GRAY (squares, broken line, vertical offset 8 units), Y/GRAY (crosses, broken line, vertical o&et 4. units) and PBjGRAY (diamonds, broken line), as a function of total effective contrast (see text). Each data point is an average of 40 observations. 95% confidence intervals, based on total variance (n = 40). larger than the data symbols are shown by error bars. The vector model (unbroken line) is reproduced for each data set. S:CES.
14
w t-
20 $0 TOTAL EFFECTIVE
4
0
I 0
1
1
I
I
10
20
30
LUMINANCE
CONTRAST
40
(%)
FIGURE 4. Contrast estimates as a function of actual 1~~~~ contrast for R/GRAY (triangles, broken tine) and the vector model (unbroken line). Each data point is an average of 40 ratings. 95% confidence intervals, based on total variance (N = 40), larger than the data symbols are shown by error bars.
(with TEC defined by the model, as above). The rating functions for the four test colors are displayed in this manner in Figs 5 (CES) and 6 (RPP), with vertical shifts to avoid overlap. The model functions {solid lines) are taken from Fig. 2 (GRAY/GRAY, EAC = 0) and are reproduced for comparison with each data set. BG/GRAY and Y/GRAY data deviate from the model in the same manner as data for R/GRAY. All three test colors show an enhancement of contrast compared to the first of the two branches of the function predicted by the model. The comparison of PB/GRAY data with the model is less clear. RPP shows a relatively large enhancement effect for PB/GRAY, but for CES the contrast estimates for PB/GRAY fit the model almost perfectly. The deviation of perceptual contrast from the model is depicted in Fig. 7 by showing the ratio of the empirical contrast to theoretical contrast, averaged across test color condition. The thick horizontal line (ratio = 1) represents a perfect match of contrast estimate to ihe model. Contrast estimates exceed tht model by a factor 1.5-2 when luminance contrast falls between roughly 5 and 15%. The data match the model at luminance contrasts above about 25%. IXSCUSSION
Conventional theories of color vision postulate scparate chromatic and achromatic (luminance) mechanisms
(e.g. Hurvich & Jameson, 1957), although a singie channel may carry chromatic and achromatic signals (Ingling & Drum, 1973; Gouras & Zrenner, 1979; Kelly, 1983). We have found that a vector model that assumes independent processing of chromatic and luminance differences cannot explain perceptual co@trast except at relatively high luminance contrasts, above 25%, where hue differences play only a minor role in visual segmentation. Where luminance contrast is relatively low, between 5 and lS%, perceptual contrast can be greater than the model by a factor of almost 2. Although the low light levels we used might limit the generality of our findings, these results suggest an interaction in the processing of spatial differences in chromaticity and luminance and are consistent with other evidence that suggests a lack of independence in the processing of color and luminance information. Switkes, Bradley and De Valois {1988) report that luminance contrast near threshold can facilitate the detection of a chromatic grating. Hilz, Huppmann- and Cavonius ( 1974), who used a small spot similar in size to our test bar, as well as gratings, found that wavelength discrimination is facilitated by luminance contrast (see also Elsner, Pokorny & Burns, 1986, who 30 1
- . 0
S:RPP
lb TOTAL
_.. --e-
20 EFFECTIVE
30
40 CONTRAST
FIGURE 6. As in Fig. S. S: RPP.
50 (%)
RGRAY
VISUAL SEGME~ATION a
2.5-
S:CES
OF A H~ER~HROMATIC
(a)
:
0
* In 2 0.52 “0 0.0
I
;;f 2.5-r
1
I
S:RPP
I
(W
8
t:
s
0.0 ,
0
1
I
I
I
10
20
30
LUMINANCE
CONTRAST
40
(%)
FIGURE 7. Average ratio of contrast estimate/model for R/GRAY, BGIGRAY, Y/GRAY, and PB/GRAY. The SE is shown by error bars. A ratio of I (thick horizontal line) indicates a perfect match between the contrast estimate and the model.
FIELD
1129
difference in luminance between the test bar and surround and therefore it is unlikely that a Bezold-Brucke hue shift contributes to contrast enhancement (Smith, Pokorny, Cohen & Perara, 1968; Hunt, 1977; Elsner ef al., 1986). Rather, it appears that the luminance border produced by the mismatch in luminance may serve to increase the saliency of chromatic contrast as a cue in figure-ground segregation. A luminance difference between a central disk of one color surrounded by an annulus of another color increases the visibility of the central disc under conditions where the border between the two fields is stabilized on the retina (Krauskopf, 1963). Under normal viewing conditions a luminance pedestal or a thin luminance ring that surrounds a chromatic test field facilitates detection of the test field (Stromeyer et al., 1989; Cole et al., 1990). Interactions between luminance borders and chromaticity differences are also basic to the gap effect (Boynton, Hayhoe & Macteod, 1977; Eskew, 1989). Luminance borders may play an important role in the visual segmentation of a field into center and surround components by increasing the difference in hue created by a spatiaf difference in chromaticity. If luminance borders have this effect, the enhancement of chromatic contrast should occur at or slightly above the luminance threshold. We plan to test this notion in future experiments by measuring luminance threshold in addition to perceptual contrast for each test bar and surround.
REFERENCES failed to find evidence of facilitation under similar conditions). We used the perceptual contrast between the two components of the test field as a global measure of the degree to which the components appear to separate from each other. In the achromatic condition (GRAY/GRAY) condition there is a change in appearance of the test bar from a surface mode to an aperture mode when luminance contrast exceeds about 25%. Although there is no theoreticai reason for a dual-branch curve relating perceptual contrast and luminance contrast, the break in the curve (Fig. 2) occurs about where there is a change in the appearance of the test bar. Kaiser et al. (1968) and Frome et al. (198 1) found that the vector model gave a better fit than a linear model to measures of border distinctness, although the fit to their data was not perfect. Since border distinctness is not a good index of the color difference between regions removed from the border (Boynton, 1978), perceptual contrast may be a better measure of overall visual segmentation. Our study shows the vector model cannot explain perceptual contrast in the 5515% luminance contrast range. In our study we did not attempt to measure the hue of the chromatic test bar, but our impression was that the hue did not change with luminance contrast. Luminance contrast of 10% requires ~0.1 log unit
Boynton, R. M. (1978). Ten years of research with the minimally distinct border. In Armington, J. C., Krauskopf, J. & Wooton, B. R. (Eds), Visualpsychophysics andphysiology (pp. 193-207). New York: Academic Press. Boynton, R. M., Hayhoe. M. M. & MacLeod, D. I. A. (1977). The gap effect: Chromatic and achromatic disc~mination as affected by field separation. Qptica Acta, 24, 159-177. Cole, C. R.. Stromeyer III, C. F. & Kronauer, R. E. (1990). Visual interactions with luminance and chromatic stimuli. Journal OJ the Optical Society of America, A 7, 128-140.
Elsner, A. E., Pokorny, J. & Burns, S. A. (1986). Chromaticity discrimination: Effects of luminance contrast and spatial frequency. Journal of the Optical Society of America, A3, 916-920. Eskew. R. 1. fI989). The gap effect revisited: Slow changes in chromatic sensitivity as affected by luminance and chromatic borders. Vision Research, 29, 717-729. Frome, F. S., Buck, S. L. & Boynton, R. M. (1981). Visibility of borders: Separate and combined effects of color differences, luminance contrast, and luminance level. Journal of the Optical Society of America,
71, 145-ISO.
Gouras, P. & Zrenner, E. (1979). Enhancement of luminance flicker by color-opponent mechanisms. Science, 20.5, 587-589. Hilt, R. L., Huppmann, G. & Cavonius, C. R. (1974). Influence of luminance contrast on hue di~riminatjon. Juurnaf of the Optical Society of America, 74, 763-766.
Hunt, R. W. G. (1977). The specification of color appearance. I. Concepts and terms. Color Research and Application, 2. 55-68. Hurvich, L. M. & Jameson, D. (1957). An opponent-process theory of color vision. Psychological Review, 64, 384-404. Hurvich, L. M. Kr.Jameson, D. (1966). The perception oflightness und darkness. Boston, N.Y.: Allyn & Bacon. Ingling, Jr, C. R. Jr Drum, B. (1973). Retinal receptive fields: Correlations between psychophysics and electrophysiology. Vision Research. 13, 1151-1163.
CHARLES
1130
E. STERNHEIM
Kaiser, P. K., Herzberg, P. A. & Boynton, R. M. (1971). Chromatic border distinctness and its relation to saturation. Vision Research. II, 935-968.
73, 742-750.
Koffka, K. (1935). Principles of Gestalt psychology. New York: Harcourt. Krauskopf, J. (1963). Effect of retinal image stabilization on the appearance of heterochromatic targets. Journal of the Optical Society
of America,
Stromeyer III, C. F., Eskew, R. E. & Kronauer, R. t. ( 19891 Chromatic facilitation by a luminance edge. Inrestigutitv Ophrhoi30. 220.
Switkes, E., Bradley, A. & De Valors. K. K. (1988) Contrast dependence and mechanisms of maskin interactions among chromatic and luminance gratings. Journal of the Optical Society oi America,
AS, 1149-l 162.
Wyszecki, G. & Stiles, W. S. (i982). Co/or sctence: Concepts methods, quantitative data and ~formufar. New York: Wiley.
and
53, 741-744.
MacAdam, D. L. (1949). Color discrimination and the influence of color contrast on visual acuity. Revue D’Optique. Theorique et Instrumentale,
P. PENN
mology and Visual Science (Suppl.),
Kelly, D. H. (1983). Spatiotemporal variation of chromatic and achromatic contrast thresholds. Journal of the Optical Society of America,
and RICHARD
28, 161-173.
Smith, V. C., Pokorny, J., Cohen, J. & Perera, T. (1968). Luminance thresholds for the Bezold-Brucke hue shift. Perception and Psychophysics, 3, 306-3 10.
Acknowledgements-We wish to thank R. M. Boynton for suggestions while this work was in progress. B. Drum. C. F. Stromeyer III provided valuable comments on a draft of this manuscript. An anonymous reviewer suggested plotting contrast estimates as a function of total effective contrast to facilitate comparison with the vector model.
Copynght
c
0042-6989/92 $5.00 + 0.00 1992 Pergamon Press Ltd
Visual Segmentation of a Heterochromatic Field Into Center/Surround Components* CHARLES
E. STERNHEIM,?
Received 4 December
RICHARD
1990; in revised form
P. PENN?
18 September
1991
The process of visual segmentation was studied by quantitatively estimating the apparent difference (i.e. the perceptual contrast) between two segments of a heterochromatic field. A chromatic test bar (Munsell 5R 6/6, 5BG 6/6, 5Y 6/6, or 5PB 6/6) was centered in an achromatic surround (Munsell N6) and the luminance contrast between the test bar and surround was varied between 0 and 39%. At equiluminance (0% luminance contrast) the chromatic border between center and surround was vague and perceptual contrast was minimal. When luminance contrast was increased slightly (5-15%) the border between center and surround was more distinct and perceptual contrast was almost two times greater than predicted by a model that combines spatial differences in chromaticity and luminance as orthogonal vectors. A luminance border may enhance perceptual contrast by increasing the salience of the center/surround hue difference. A vector model is consistent with perceptual contrast only at relatively high luminance contrasts, where hue differences play a relatively minor role in visual segmentation. Perceptual contrast
Luminance contrast
Figure-ground
INTRODUCTION Spatial differences in luminance and chromaticity combine to influence the visibility of an object in a scene. For example, the outline of a blue figure on a gray background depends upon the luminance difference between the two visual areas. KofIka (1935) observed that without luminance contrast there is a tendency for a blue figure to lose sharpness and with steady fixation to fade from view. Perception becomes stable only if the figure and ground are non-equiluminant. The change in visibility of a chromatic field as a function of luminance contrast raises the question of whether spatial differences in chromaticity and luminance are processed independently. The aim of the present study was to scale the combined effect of chromatic and luminance contrast and to test a vector model that assumes independent processing of chromatic and luminance differences. The vector model was developed by MacAdam (1949) to account for visual acuity measured with chromatic Landolt Cs against a neutral background. The same mode1 has been used by Kaiser, Herzberg and Boynton (1971) and Frome, Buck and Boynton (1981) to explain the distinctness of a border between two heterochromatic fields. In our experiment a chromatic test bar was positioned in the center of an achromatic surround *A preliminary report based on this study was presented at the 1987 OSA Meeting, Rochester, N.Y. tDepartment of Psychology, University of Maryland, College Park, MD 20742, U.S.A.
Segregation
and subjects scaled perceptual contrast. Chromatic contrast at equiluminance was converted into equivalent achromatic contrast (EAC) and combined with actual luminance contrast by orthogonal vector addition. Results show that in the range of luminance contrast between roughly 5 and 15% perceptual contrast is enhanced compared to the vector model. This suggests that spatial differences in chromaticity and luminance may interact to improve the segmentation of a field into center and surround components. The data are consistent with the results of other research that support the notion that chromatic contrast may be enhanced by luminance borders (Stromeyer, Eskew & Kronauer, 1989; Cole, Stromeyer & Kronauer, 1990).
METHOD Stimulus and apparatus
The test field consisted of a test bar located in the center of a neutral gray surround (Fig. 1, right side). Munsell (1976) papers were positioned so that each component of the test field could be illuminated independently. Munsell paper N6, containing a rectangularshaped aperture, functioned as the surround and was placed 56.5 in. from the subject. Munsell papers 5R 6/6, 5BG 6/6, 5Y 6/6, 5PB 6/6, or N6 were placed 6 in. directly in back of the aperture. Light reflected from the rear paper passing through the aperture constituted the test bar and traveled in the same direction as light reflected from the surround into the subject’s eye. When N6 was placed behind the aperture (N6/N6 condition)
1125
1126
CHARLES E. STERNHEIM and RICHARD P. PENN
An experimental session began with 8 min ot’ dark adaptation followed by 2 min of light adaptation to the test field. Adaptation was controlled almost entirely by the surround since it occupied the largest portion of the test field. After the adaptation period the subject used a minimally distinct border criterion to adjust the luminance of the test bar to match the luminance of the surround. The average of five independent settings was used to determine ~uiluminancc. In the trials that followed the experimenter varied Comparison luminance contrast in a mixed order. The subject was given no feedback about luminance contrast. On each trial the subject was instructed to scan acrass the test field for 3-5 set, avoiding steady fixation. The subject’s task was to rate the perceptual contrast contained in the test tieid by considering the degree to which the test bar appeared to separate from the surround. The FIGURE 1. The test &Id, on the right, consists of a test bar eaciased comparison bar viewed in darkness was used as a by a rectangular surround (not to scat+ The compatisonfield, on the standard and given a value of “IO”. The subject rated left, is a bar equal in size to the test bar in a dark surround. the test bar with any positive number, including any decimal number. Subjects were well practiced in making the test bar was unseen when the two grays were made numerical ratings before data were collected. Twelve equiluminant. A test bar was visible at equiiuminance levels of luminance contrast were presented 10 times each in every session. Viewing conditions were replicated only when the chroma~~ty of the rear paper differed in separate sessions 4 times in a mixed order. from the chromaticity of the surround.* A comparison In order to rate perceptual contrast, the subject bar (Munsell paper N6) of fixed huuinance was attended to the difference in appearance between the positioned to the left of the test field (Fig. 1) at the same area covered by the test bar and the area covered by the distance from the subject. surround. For example, a gray (N6) test bar with Munsell papers were illuminated by light originating luminance contrast in the range l-5% appeared lighter from a tungsten-halogen source flltered through a combination of four Kodak filters (82C, 82C, 82A, than the gray (N6) surround. However, the difference in 8C). The fWred light was 6060 K and the CfE 1931 lightness was very slight and the test bar and surround chromatic&y coordinates (x, y) corresponding to the practically merged with each other. Perceptual contrast Munseli papers were as follows: 5R 6/6 (0.420, 0.363); was typically rated under 0.5. In comparison, a red f5R SBG 6/6 (0.266, 0.375); 5Y 6/6 (0.431, 0.452); 5PB 6/(i) test bar with 22.7% luminance contrast looked considerably lighter and redder than the surround. Since (0.272, 0.313); and N6 (0.333, 0.364). The surround and comparison bar were fixed at 1.43 log cd/m*. The the test bar appeared to separate more clearly from the luminance of the test bar was varied from equiluminant surround, perceptual contrast was typically rated close to the surround and up by turning a circular neutral to 10.0. density wedge located at a focal plane of the test bar optical pathway. Twelve levels of kminance contrast between test bar and surround were produced: 0 (equihtminance), 1.4, 2.8, 4.6, 9.2, 13.7, 18.2, 22.7, 27.0, 31.2, 35.3, and 39.3%. [Contrast = (I,,, - &,J(L,, + I+,&_] Color temperature, chromaticity, and luminance were calibrated in situ with a Minolta Chroma Meter (~S-~~). Procedure The subject, with a chin rest to help stabilize head position, viewed the test field and the comparison bar monocularly without an artifkiai pupil or achr~at~ng lens, The outer edges of these fitds aided in accommodation. of Mumull papars to form the test &Id in this mamar is similar to the experimental set-up that was used by Hess and Pretori (1894, as described by Hurvich dr Jameson, 1966) to study
*The positioning
brightness contrast.
The two authors served as subjects. They each have normal color vision and normal acuity with correction. CES had previously participated as a subject in psychophysical experiments. Since this was the first psychophysical experiment for RPP, he received extensive training in prelimina~ work where the procedures of this experiment were refined, Both subjects had knowledge of the experimental questions, but they were kept uninformed about their results during an experimental session. REWLTS
The contrast rating functions for GRAY/GRAY (N6/N6) given in Fig. 2 are based on averages from four sessions. There were only minor deviations in the results of individual sessions. Luminance contrast below about 25% is perceived as a lightness difference between center and surround, whereas luminance contrast above
VISUAL SEGMENTATION
:
‘4 - ORAYIGRA~ 12- . S:RPP
s F
lo-
:
1127
FIELD
+
EGGRAY
+
YlGRAY
-
P&GRAY
8-
ii) d
6-
g
4-
8
OF A HETEROCHROMATIC
20
O!
0
10 LUMINANCE
20 CONTRAST
30
40
(K)
FIGURE 2. Contrast estimates in the GRAY/GRAY condition. Each data point is an average of 40 ratings. 95% confidence intervals, based on total variance (n = 40), larger than the data symbols am shown by
error bars. The broken lines connect the data points.
approx. 25% is perceived primarily as a differece in brightness.* This change in appearance from a surface
mode to an aperture mode occurs near where there is a break in the contrast rating function. The dual-branch functions that best fits the data of each observer are shown by solid lines. If P represents the contrast estimate and L represents the luminance contrast, the functions for CES’s data [Fig. 2(a)] are log P = 2.43 log L - 2.50 for the lower branch and log P = 0.91 log L - 0.38 for the upper branch. The functions for RPP [Fig. 2(b)] are log P = 2.99 log L - 3.42 for the lower branch and log P = 1.48 log L - 1.27 for the upper branch. The broken lines in this figure connect individual data points. The perceptual contrast of an equiluminant chromatic test bar varies markedly with test bar chromaticity (Fig. 3). R/GRAY has more contrast than BG/GRAY, followed in order by Y/GRAY, and PB/GRAY. The contrast rating functions increase with luminance contrast and the four functions converge at the higher levels, where the test bar is clearly brighter than the surround regardless of test bar chromaticity. The EAC of a chromatic test bar is the luminance contrast which would produce, in the GRAY/ GRAY (N6/N6) condition, the same rating that
*Luminance contrast between 1 and 5% is above threshold, but the contrast estimates in this range are too low to plot clearly above the horizontal axis.
0
10 LUMINANCE
20 CONTRAST
30
40
(%)
FIGURE 3. Contrast estimates for R/GRAY (triangles), BG/GRAY (squares), Y/GRAY (pluses), and PB/GRAY (diamonds). Each data point is an average of 40 ratings. The EAC values are as follows: R/GRAY (16.3 and 18.2% for CES and RPP, respectively);
BG/GRAY (13.7%, 12.8%), Y/GRAY (7.8%. 7.2%); PB/GRAY (2.4%, 3.0%).
the equiluminant chromatic test bar produced. If P represents the contrast estimate, EAC is obtained by using P =f(L) (L being the luminance contrast and f is the dual-branched function of Fig. 2). An EAC for each chromatic test bar is given in the legend of Fig. 3. The vector model predicts that P,, =f [(EAC' + /2)o.5]. For example, the average rating given by CES to an equiluminant R/GRAY test bar is 2.8 [Fig. 3(a) triangles, luminance contrast = 01. The EAC of R/GRAY for CES is 16.3%. If 9.2% luminance contrast is introduced for this test color, the total effective contrast [TEC = (EAC’ + 12)o.s]is 18.7% and P,, is 3.9. Empirical estimates of contrast assigned to R/GRAY conform with the model (Fig. 4), except for a limited range of luminance contrast, between roughly 5 and 15% where contrast estimates are comparatively high. Although the disagreement with the model is not very marked, enhancement of contrast at moderate levels of luminance contrast appears to be a reliable effect for both observers. Since EAC varies with test bar chromaticity, the shapes of the theoretical contrast rating functions, as plotted as in Fig. 4 with actual luminance contrast as the abscissa, are expected to vary. However, theoretical functions are expected to have the same shape if contrast estimates are plotted against total effective contrast
1128
CHARLES
E. STERNHEIM
and RICHARD 30
14
-I
P. PENN
S:CES
12 10 8 PEYGRAY
6
MODEL
4
-
MODEL
0 0
12
z
10
3
6
& d 5
6
8
2
30 40 CONTRAST (%)
50
FIGURE 5. Contrast estimates for R/GRAY (triangles. broken line, vertical offset 12 units), BG/GRAY (squares, broken line, vertical offset 8 units), Y/GRAY (crosses, broken line, vertical o&et 4. units) and PBjGRAY (diamonds, broken line), as a function of total effective contrast (see text). Each data point is an average of 40 observations. 95% confidence intervals, based on total variance (n = 40). larger than the data symbols are shown by error bars. The vector model (unbroken line) is reproduced for each data set. S:CES.
14
w t-
20 $0 TOTAL EFFECTIVE
4
0
I 0
1
1
I
I
10
20
30
LUMINANCE
CONTRAST
40
(%)
FIGURE 4. Contrast estimates as a function of actual 1~~~~ contrast for R/GRAY (triangles, broken tine) and the vector model (unbroken line). Each data point is an average of 40 ratings. 95% confidence intervals, based on total variance (N = 40), larger than the data symbols are shown by error bars.
(with TEC defined by the model, as above). The rating functions for the four test colors are displayed in this manner in Figs 5 (CES) and 6 (RPP), with vertical shifts to avoid overlap. The model functions {solid lines) are taken from Fig. 2 (GRAY/GRAY, EAC = 0) and are reproduced for comparison with each data set. BG/GRAY and Y/GRAY data deviate from the model in the same manner as data for R/GRAY. All three test colors show an enhancement of contrast compared to the first of the two branches of the function predicted by the model. The comparison of PB/GRAY data with the model is less clear. RPP shows a relatively large enhancement effect for PB/GRAY, but for CES the contrast estimates for PB/GRAY fit the model almost perfectly. The deviation of perceptual contrast from the model is depicted in Fig. 7 by showing the ratio of the empirical contrast to theoretical contrast, averaged across test color condition. The thick horizontal line (ratio = 1) represents a perfect match of contrast estimate to ihe model. Contrast estimates exceed tht model by a factor 1.5-2 when luminance contrast falls between roughly 5 and 15%. The data match the model at luminance contrasts above about 25%. IXSCUSSION
Conventional theories of color vision postulate scparate chromatic and achromatic (luminance) mechanisms
(e.g. Hurvich & Jameson, 1957), although a singie channel may carry chromatic and achromatic signals (Ingling & Drum, 1973; Gouras & Zrenner, 1979; Kelly, 1983). We have found that a vector model that assumes independent processing of chromatic and luminance differences cannot explain perceptual co@trast except at relatively high luminance contrasts, above 25%, where hue differences play only a minor role in visual segmentation. Where luminance contrast is relatively low, between 5 and lS%, perceptual contrast can be greater than the model by a factor of almost 2. Although the low light levels we used might limit the generality of our findings, these results suggest an interaction in the processing of spatial differences in chromaticity and luminance and are consistent with other evidence that suggests a lack of independence in the processing of color and luminance information. Switkes, Bradley and De Valois {1988) report that luminance contrast near threshold can facilitate the detection of a chromatic grating. Hilz, Huppmann- and Cavonius ( 1974), who used a small spot similar in size to our test bar, as well as gratings, found that wavelength discrimination is facilitated by luminance contrast (see also Elsner, Pokorny & Burns, 1986, who 30 1
- . 0
S:RPP
lb TOTAL
_.. --e-
20 EFFECTIVE
30
40 CONTRAST
FIGURE 6. As in Fig. S. S: RPP.
50 (%)
RGRAY
VISUAL SEGME~ATION a
2.5-
S:CES
OF A H~ER~HROMATIC
(a)
:
0
* In 2 0.52 “0 0.0
I
;;f 2.5-r
1
I
S:RPP
I
(W
8
t:
s
0.0 ,
0
1
I
I
I
10
20
30
LUMINANCE
CONTRAST
40
(%)
FIGURE 7. Average ratio of contrast estimate/model for R/GRAY, BGIGRAY, Y/GRAY, and PB/GRAY. The SE is shown by error bars. A ratio of I (thick horizontal line) indicates a perfect match between the contrast estimate and the model.
FIELD
1129
difference in luminance between the test bar and surround and therefore it is unlikely that a Bezold-Brucke hue shift contributes to contrast enhancement (Smith, Pokorny, Cohen & Perara, 1968; Hunt, 1977; Elsner ef al., 1986). Rather, it appears that the luminance border produced by the mismatch in luminance may serve to increase the saliency of chromatic contrast as a cue in figure-ground segregation. A luminance difference between a central disk of one color surrounded by an annulus of another color increases the visibility of the central disc under conditions where the border between the two fields is stabilized on the retina (Krauskopf, 1963). Under normal viewing conditions a luminance pedestal or a thin luminance ring that surrounds a chromatic test field facilitates detection of the test field (Stromeyer et al., 1989; Cole et al., 1990). Interactions between luminance borders and chromaticity differences are also basic to the gap effect (Boynton, Hayhoe & Macteod, 1977; Eskew, 1989). Luminance borders may play an important role in the visual segmentation of a field into center and surround components by increasing the difference in hue created by a spatiaf difference in chromaticity. If luminance borders have this effect, the enhancement of chromatic contrast should occur at or slightly above the luminance threshold. We plan to test this notion in future experiments by measuring luminance threshold in addition to perceptual contrast for each test bar and surround.
REFERENCES failed to find evidence of facilitation under similar conditions). We used the perceptual contrast between the two components of the test field as a global measure of the degree to which the components appear to separate from each other. In the achromatic condition (GRAY/GRAY) condition there is a change in appearance of the test bar from a surface mode to an aperture mode when luminance contrast exceeds about 25%. Although there is no theoreticai reason for a dual-branch curve relating perceptual contrast and luminance contrast, the break in the curve (Fig. 2) occurs about where there is a change in the appearance of the test bar. Kaiser et al. (1968) and Frome et al. (198 1) found that the vector model gave a better fit than a linear model to measures of border distinctness, although the fit to their data was not perfect. Since border distinctness is not a good index of the color difference between regions removed from the border (Boynton, 1978), perceptual contrast may be a better measure of overall visual segmentation. Our study shows the vector model cannot explain perceptual contrast in the 5515% luminance contrast range. In our study we did not attempt to measure the hue of the chromatic test bar, but our impression was that the hue did not change with luminance contrast. Luminance contrast of 10% requires ~0.1 log unit
Boynton, R. M. (1978). Ten years of research with the minimally distinct border. In Armington, J. C., Krauskopf, J. & Wooton, B. R. (Eds), Visualpsychophysics andphysiology (pp. 193-207). New York: Academic Press. Boynton, R. M., Hayhoe. M. M. & MacLeod, D. I. A. (1977). The gap effect: Chromatic and achromatic disc~mination as affected by field separation. Qptica Acta, 24, 159-177. Cole, C. R.. Stromeyer III, C. F. & Kronauer, R. E. (1990). Visual interactions with luminance and chromatic stimuli. Journal OJ the Optical Society of America, A 7, 128-140.
Elsner, A. E., Pokorny, J. & Burns, S. A. (1986). Chromaticity discrimination: Effects of luminance contrast and spatial frequency. Journal of the Optical Society of America, A3, 916-920. Eskew. R. 1. fI989). The gap effect revisited: Slow changes in chromatic sensitivity as affected by luminance and chromatic borders. Vision Research, 29, 717-729. Frome, F. S., Buck, S. L. & Boynton, R. M. (1981). Visibility of borders: Separate and combined effects of color differences, luminance contrast, and luminance level. Journal of the Optical Society of America,
71, 145-ISO.
Gouras, P. & Zrenner, E. (1979). Enhancement of luminance flicker by color-opponent mechanisms. Science, 20.5, 587-589. Hilt, R. L., Huppmann, G. & Cavonius, C. R. (1974). Influence of luminance contrast on hue di~riminatjon. Juurnaf of the Optical Society of America, 74, 763-766.
Hunt, R. W. G. (1977). The specification of color appearance. I. Concepts and terms. Color Research and Application, 2. 55-68. Hurvich, L. M. & Jameson, D. (1957). An opponent-process theory of color vision. Psychological Review, 64, 384-404. Hurvich, L. M. Kr.Jameson, D. (1966). The perception oflightness und darkness. Boston, N.Y.: Allyn & Bacon. Ingling, Jr, C. R. Jr Drum, B. (1973). Retinal receptive fields: Correlations between psychophysics and electrophysiology. Vision Research. 13, 1151-1163.
CHARLES
1130
E. STERNHEIM
Kaiser, P. K., Herzberg, P. A. & Boynton, R. M. (1971). Chromatic border distinctness and its relation to saturation. Vision Research. II, 935-968.
73, 742-750.
Koffka, K. (1935). Principles of Gestalt psychology. New York: Harcourt. Krauskopf, J. (1963). Effect of retinal image stabilization on the appearance of heterochromatic targets. Journal of the Optical Society
of America,
Stromeyer III, C. F., Eskew, R. E. & Kronauer, R. t. ( 19891 Chromatic facilitation by a luminance edge. Inrestigutitv Ophrhoi30. 220.
Switkes, E., Bradley, A. & De Valors. K. K. (1988) Contrast dependence and mechanisms of maskin interactions among chromatic and luminance gratings. Journal of the Optical Society oi America,
AS, 1149-l 162.
Wyszecki, G. & Stiles, W. S. (i982). Co/or sctence: Concepts methods, quantitative data and ~formufar. New York: Wiley.
and
53, 741-744.
MacAdam, D. L. (1949). Color discrimination and the influence of color contrast on visual acuity. Revue D’Optique. Theorique et Instrumentale,
P. PENN
mology and Visual Science (Suppl.),
Kelly, D. H. (1983). Spatiotemporal variation of chromatic and achromatic contrast thresholds. Journal of the Optical Society of America,
and RICHARD
28, 161-173.
Smith, V. C., Pokorny, J., Cohen, J. & Perera, T. (1968). Luminance thresholds for the Bezold-Brucke hue shift. Perception and Psychophysics, 3, 306-3 10.
Acknowledgements-We wish to thank R. M. Boynton for suggestions while this work was in progress. B. Drum. C. F. Stromeyer III provided valuable comments on a draft of this manuscript. An anonymous reviewer suggested plotting contrast estimates as a function of total effective contrast to facilitate comparison with the vector model.