I awn Rc.F. Vol. IS. pp. 917-929. Perpamon Press 1975. Prmted m Great Britain

COLOR AND BRIGHTNESS CONTRAST EFFECTS AS A FUNCTION OF SPATIAL VARIABLES’ E. WILLIAM YUND’ and JOHN C. ARMINGTON Department of Psychology, Northeastern University, Boston, Massachusetts 02115. U.S.A. (Received

26 December

1973; in

revisedform 3 December

1974)

Abstract-The contrast effect was measured at the center of a circular center-surround stimulus display for three color combinations and one (non-colored) brightness combination. For each combination, the magnitude of the contrast effect was determined for a series of different center and surround sizes using consecutive matching procedures. The amount of contrast was influenced by the dimensions of the display in the expected way; the effect on the center was greatest when the surround was large and the center was small. The contrast magnitudes were plotted against four theoretically selected abscissae: (1) surround width, R, - R,; (2) surround area, R: - Ri; (3) area ratio, (RI - R$/Ri; and (4) edge-distance expression I/RI - l/R,, where R, is the radius of the central test area and R, is the outer radius of the surround. For each color combination, both graphical and correlational analyses demonstrated that contrast magnitude is more closely related to the edge-distance expression than to the other spatial expressions. This result provides support for an edge-distance model based on two assumptions: (1) that edges in the stimulus, and edge detectors in the visual system. are the important determinants of color and brightness; and (2) that edges closer to a point contribute more to the determination of the color and brightness at that point than do edges that are further away.

The perceived color and brightness of one area of a stimulus depends not only upon the spectral energy distribution of the light coming into the eye from that area, but also upon the light coming from adjacent and surrounding areas. Such other areas most often cause the particular area under consideration to appear more different from them than it would if viewed in isolation. A gray surrounded by white appears darker; the same gray surrounded by black appears lighter. A yellow-green surrounded by green appears more yellowish; the same yellow-green surrounded by yellow appears more greenish. Since these effects seem to enhance differences, they have been called contrast effects or simultaneous contrast effects -to differentiate them from the successive contrast effects which occur when stimuli are presented in a temporal sequence to the same retinal location. It is also important not to confuse contrast effects, as described above, with objective contrast, a measure of the luminance difference between two areas. In this report, the term “contrast” will be used only to refer to perceived simultaneous contrast effects. It is well known that color and brightness contrast effects vary with the spatial parameters of the visual stimulus, but surprisingly little effort has been directed towards the discovery of the relevant stimulus dimensions. At present, the relationship between contrast effect magnitudes and spatial variables can be ‘The research described in this paper was supported in part by USPHS grant EYOO589; the neurophysiological results described here were supported in part by USPHS grant EYOO014to R. De Valois. The first author has been supported by USPHS predoctoral fellowship GM36258. postdoctoral fellowship EY36248 and the Dept. of Medicine and Surgery, Veterans Administration. Neurophysiology-Biophysics Address: ’ Present Research Laboratories, Veterans Administration Hospital. 150 Muir Road. Martinez, California 94553, U.S.A. 917

described only in approximate quantitative terms. Wallach (1948), Diamond (1955), Stevens (1967) and Heinemann (1972) have demonstrated that brightness contrast effects increase as the size of the inducing field increases. In all of these experiments only the size of the inducing field was varied; the size of the test field was held constant. Kinney (1962) studied color contrast as a function of inducing and test field size. The overall stimulus size was kept constant, with different proportions of this area occupied by the test and inducing fields. The amount of induced color increased as the proportion of the stimulus occupied by the inducing field increased. It has also been shown, for both brightness (Leibowitz, Mote and Thurlow, 1953; Dunn and Leibowitz, 1961; Mackavey, Bartley and Casella, 1961) and color (Jameson and Hurvich, 1961; Oyama and Hsia. 1966), that contrast effects decrease as the separation between test and inducing fields increases and that the greatest contrast effects occur in small test fields (Marsden, 1969). In summary, contrast effects have their greatest magnitudes when the test field is small and the inducing field is large and is in contact with and completely surrounding the test field. Essentially all of the many proposed explanations for contrast effects seem to be consistent with the relatively imprecise quantitative data now available. In facf data from experiments featuring extensive variation of spatial parameters are needed to test the current models. The purpose of this study was to begin the task of gathering such data and to apply these newly acquired data to test several of the possible models for contrast effects. METHODS Sritnuli The stimuli presented to the observer during a single trial of the color contrast experiment are illustrated in

91s

E. 1 5 set

o-5

I 5 see

WILLMM

YUHD and

I 5 see

‘&:’ contrast S,IITI”I”S

Motchtnq stimulus

Fig. 1. Spatial and temporal configuration of stimuli for color contrast experiments. Center and surround sizes ranged from 1” to 20”. The contrast and matching stimuli were presented for 1.5 set in succession, separated by 0.5 sec. In the brightness contrast experiment, there was no surround on the matching stimulus. For all of the results reported here. both stimuli were delivered to both eyes. Fig. 1. One trial consisted of a 1.5~set presentation each of the contrast and the matching stimulus separated by 0.5 sec. Further description of the temporal and chromatic aspects of these stimuli will be given subsequently (see Procedures). Both the contrast and the matching stimuli had a circular center-surround configuration and were delivered binocularly. The center and surround for each stimulus was presented to both eyes. The resulting image appeared to the observer to be directly in front of him at a distance of eight inches. In the brightness contrast procedure, the matching stimulus had no surround; in the terminology of Fig. 1, there was no &. The subject was asked to fixate on the center of the stimulus and to base his judgments of the color and brightness of the stimuli on the appearance at the center.

JOHN C. ARMINGTON

quickly and easily. A set of pairs of each of the changeable screens was made. the center diameters going in 11 equal steps from 15 to 19’ and the surround outer diameters in 11 steps from 3. to 20.5’. It should be noted that the smallest center diameter used here is somewhat larger than the test fields used in many brightness contrast experiments. This size limit was chosen for two reasons: (1) to be more consistent with stimulus sizes used m color contrast experiments; and (2) to avoid possible confounding of the larger distance brightness contrast effects with Mach bands (Bekisy. 1968). Forty center-surround size combinations were used for each of the three color combinations. The center and outer surround radii are given in Table 1. Filters and monochromator settings for the three color combinations are

”

M

1

Apparatus

The optical system for the color experiments is shown in Fig. 2. The overall assembly consisted of two separate systems, one providing the light for the center of the stimuli of Fig. 1 and the other that for the surround. The beams of the center system are indicated by the standard center line (--) and those of the surround by a modified center line (. -). The surround system began at sources Sl and S2, GE TlO lamps (18 A, 6 V, SR-8 filament, approximate color temperature 2500°K). The light from these sources was filtered (Fl, F2, F3) and the filament image focused in the plane of the sectored disk SD, which provided the temporal control of the stimuli. After the sectored disk, the beams were mixed and/or split by beamsplitter Bl. The resulting two beams illuminated the surround screens SC3 and SC4. A Bausch and Lomb high intensity grating monochromator (catalog No. 33-86-02) with its own tungsten source and source S3 (the same lamp as S2, the other side of the filament) provided light for the center system. The light from S3 was flltered (F4, FS). Then that filament image and the monochromator exit slit image were focused in the plane of SD. After the sectored disk, these beams were mixed and/or split by beamsplitter B2. The resulting two beams illuminated the center screens SC5 and SC6. Magnesium oxide coated screens SC3-6 were viewed directly by the observer through mirrors Ml and M2. These light paths, indicated by double solid lines, contained no lenses. The result was a fused stereoscopic image of a circular center-surround stimulus directly in front of the observer at a distance of 8 in. Since none of the observers (all under 30 yr of age) reported any difficulty in keeping the image in sharp focus at this distance, no corrective lenses were used. Screens SC1 and SC2 were painted flatblack. The hole in these screens determined the outer diameter of the surround. Screens SC3 and SC4 provided the surround and the hole in them determined the diameter of the center. The central area was provided by screens SC5 and SC6 which were permanently mounted. Screen holders for SC1-4 permitted these screens to be changed

Fig. 2. Optical system for the color contrast experiments. This system had the capacity for delivering the stimuli of Fig. 1 binocularly (both center and surround to both eyes). monocularly (center and surround to one eye and no stimulus to the other) and dichoptically (center to one eye and surround to the other]. White light sources (Sl, S2. S3) and the monochromator (M) are near the top of the figure and the observer’s eyes (LE. RE) are at the bottom. The beams which illuminated the center screens (SC5. SC6). providing both the contrast and the matching centers of Fig. 1, are indicated by the standard center line (- -). The beams which illuminated the surround screens (SC3, SC4). providing both the contrast and the matching surrounds, are indicated by the modified center 1.The size of the hole in the surround line (-. screens determined the srze of the center; the hole in the. black background screens (SCl, SC2) determined the position of the outer edge of the surround. Bl and B2 are beamsplitters which mixed and/or split beams to illuminate the two surround and center screens respectively. Fl was the neutral density filter controlling the luminance of the matching surround. F2 and F3 controlled the color and luminance of the contrast surround. F4 and F5 controlled the color and luminance of the contrast center. The sectored disc (SD) determined the temporal sequence of the stimuli.

919

Color and brightness contrast effects Table 1. Center and outer surround radii for the stimuli u$ed for the color experiments

Stimulus

Outer surround radius

Center radius

AB* BC DE* FG HI* IJ JK’ KL* AC FH* GI JL* AD* DG HK IL* AE BF GK* HL

10.62 9.75 8-01 6.24 447 3.58 2.68 I,79 10.62 6.24 5-36 2.68 10.62 8.01 4.47 358 10.62 9.75 5.36 4.47

9.75 8.88 7‘13 5-36 3.58 2.68 1.79 0.90 8.88 4.47 3.58 @90 8.01 5.36 1.79 0.90 7.13 6.24 1.79 0.90

Outer surround radius

Stimulus AF* CH* EJ FK GL* AG CI DJ FL AH* BI DK” EL* AI DL AJ* CL* AK* BL AL*

Center radius 6.24 4.41 2.68 I .19 090 5.36 3.58 2.68 0.90 4.47 3.58 I.79 O-90 3.58 0.90 2.68 0.90 1.79 0.90 090

10-62” 8.88 7.13 6.24 5.36 10.62 8.88 8.01 6.24 IO.62 9.75 8.01 7.13 1062 8.01 IO.62 8.88 10.62 9.75 I@62

* Those also used in the brightness experiment.

Table 2. Surround and center color and luminance specifications Color combination II

I Contrast surround Color filter* Screen luminance (It-L) Contrast center Color filter(s)* Screen luminance (ft-L) Matching surround Screen luminance (ft-L) Lamp current (amps) Monochromatic light Wavelength (nm) Screen luminance (ft-L/mm slit width)

III

5433 @0163

4010 0.0304

3480 0.1250

~.~ 0.0362

4015 PO607

3484 0.3983

0.0655 16

03550 15

0.3550 15

565 0017

600 0025

560 0.03 1

* Corning glass color filters; Corning identification numbers listed in table.

given in Table 2. Table 3 gives the approximate color names for the center and surround of the contrast stimulus for the three color combi~tions. The optical system for the brightness experiments was that of Fig. 2 with three modifications: (1) the monochromator was not used; (2) mirror Ml2 was moved to the left so that the light beam leaving lens Ll5 was reflected into lens L12; and (3) a density wedge was inserted between lens L12 and beamsplitter 82. The 20 center-surround size combinations used in the brightness contrast experiment are indicated by an asterisk in Table 1. One combination of Iuminances was used: contrast surround. 0.135 ft-L; contrast center. 0,175 R-L. The lamps were run at 16A. Procedure

The aims of the experimental procedure was to measure the ~gnitude of the contrast effect for each pair of centersurround sizes. For color, this was done by the method of constant stimuli in a series of successive comparison trials. A single trial is illustrated in Fig. 1: The contrast stimulus, with a colored center and surround, was presented for 1.5 set; there was a 05-set period of darkness; “At.

I_(--R;9--r

the matching stimulus, with a colored center and a white surround, was presented for 1.5 set; there was a 15-set period of darkness. During this second dark period the observer indicated the relative colors of the contrast and matching centers by saying which hue appeared to be closer to the long wavelength end of the spectrum. The white surround on the matching stimulus serving to minimize brightness differences, was adjusted to make the contrast and matching centers appear approximately equally bright to the nearest O-3 density unit (filter Fl) as judged

Table 3. Approximate color names for center and surround colors of contrast stimuli Color combination I I::

Center color

Surround color

Blue-green Yellow-green Orange

Blue Green Red

See Table 2 for specification luminances.

in terms of filters and

910

E. WILLIAMYUND and

by the principal experimenter. Control expertments measuring contrast magnitudes with the matching surround 0.3 brighter and 0.3 darker than the values used. demonstrated that the exact luminance of the matching surround had no signi~cant effect upon the measured contrast magnitudes. The light for the matching center was the same light used for the contrast center plus some amount of an appropriate monochromatic light. The wavelength of monochromatic light was chosen by trial and error. beginning with a wavelength complimentary to the surround. such that when added to the matching center it tended to mimic the contrast effect of the surround on the contrast center. This wavelength was held constant for any given color combination. Thus. a yellow-green, 56.5nm, monochromatic light was used with the blue surround blue-green center. color combination I, since the effect of the contrast surround is to make the center appear less blue and more green.

Both the contrast produced by the surround in the contrast stimulus and the monochromatic light added to the center of the matching stimulus desaturated the center. This was in addition to producing the intended color shift. In all three color combinations used here, the contrast in the test stimulus caused more desaturation of the center than the monochromatic light added to the matching stimutus. The observer’s task seemed easier when the matching stimulus was not so desaturated. The observers were instructed to disregard saturation and brightness differences as much as possible and make their judgment solely on the basis of the hue aspects of the stimuli. With some experience, they reported little difficulty in doing so. As indicated above, the method of constant stimuli was used to determine the amount of monochromatic light needed in the matching stimulus to equal the effect of the surround in the contrast stimulus. The intensitv of monochromatic light was varied by adjusting the width of the monochromator entrance slit. Any change in the bandwidth of the monochromatic light is deemed negligible for the purposes of this experiment. The procedure for determining the contrast magnitude for one centersurround sixe was as follows: (1) the room lights were turned off (room lights were “daylight white” fluorescent producing an average wall luminance of I.66 ft-L); (2) a variable number of range-finding trials was done as needed to find the approximate amount of monochromatic light necessary to match the contrast: (3) 30 trials were done using six amounts of mon~hromatic light in the immediate range of the approximate match. using 0.4 mm stit width steps which corresponded to 00043, OGO62and @0078ft-L luminance steps for color combinations I, II and III, respectively (see Calibration section); (4) there was a 1%~ break; (5) 30 more trials were done; (6) the room lights were turned on. The 60 trials consisted of 10 repetitions of each of the six levels of monochromatic light. The magnitude of the contrast effect was taken to be the luminance of the monochromatic light added such that the observer would report it to be less than the contrast effect on 50 per cent of the trials and more than the contrast effect on 50 per cent of the trials. This was determined by dividing the interval in which the 50 per cent point fell, proportionally-50 per cent is halfway in an interval with 40 and 60 per cent end points and is one-sixth of the way for 40 and 100 per cent end points. Eight such determinations were made per I-hr session A single random order of the 40 sizes was used with different observers beginning at different places. It was necessary for all observers to repeat the sequence of 40 sizes two or three times to eliminate an initial large session-to-session variability. The last repetition was used in the analysis. The magnitude of the brightness contrast effect was measured by the method of adjustment. The timing for a single

JOHN C.

ARMIXGTOS

compartson trial was the same as for color (Fig. I. top]: The contrast stimulus was presented for I.5 SK; there was a 05-set period of darkness: the matching center was present for I.5 set: there was a l+sec pertod of darkness. During this second dark period the observer indicated how to adjust the matching center to make the two centers equally bright: the experimenter then adjusted the densit! wedge accordingly. The procedure was continued until the observer indicated that a match was made. The observers were encouraged to be critical in their acceptance of a match. to try a little brighter and a little darker to he sure to find the closest possible setting. The magnitude of contrast is expressed as the density setting of the wedge which produced the match. One determination proceeded as follows: Room lights were turned OK: the match was made: then room lights were turned on again. Ten sizes were done in a I-hr session. Two matches were made for each of the sizes in a session. one starting with the matching center brighter and one with it darker than the test center. Each observer made four matches for each size combination used. two of which were taken at one session and two at another. The average of the four is taken as the relative magnitude of the contrast of that observer for that size. Caiibrarim

The luminance of the stimuli was measured with a Spectra spot meter. With the spot meter taking the place of the observer. the luminance of the screens SC3-6 as reflected in mirrors Ml and M2 (double line light paths of Fig. 2) was measured for each of the surfaces viewed by the subject. These values for the color stimuli are included in Table 2. The measurement for the monochromatic light was repeated for a number of entrance slit widths (@5, 1.0. 2.0, 3.0. 4.0, 5.0 and 6.0 mm) and the luminance was found linear with slit width up to 5.0 mm. Since no values of slit width greater than 5.0 mm were found for any size and color combination, a proportionality factor could be calculated for each of the three color combinations. This factor was multiplied by the slit width to obtain the luminance of monochromatic light added, in foot-lamberts. The luminances for the brightness experiment were 0.135 ft-L for the contrast surround. O.l75ft-L for the contrast center and 0.155 R-L for the matching center with the density wedge set at 0.3. Subjects In all, four observers participated. All had normal color vision as tested by the Ishihara isochromatic plates. Three (LMB. PK. MAY) were naive with respect to possible models for contrast effects. Results were collected from three observers for color-combination I (LMB. MAY. EWY). three observers for color combination II (LMB. MAY, EWY), two observers for color combination III (LMB, EWY) and three observers for the brightness combination (LMB, MAY, EWY). RESULTS

The magnitudes of the contrast effects for each center-surround size combination, averaged across subjects within each color and brightness combination. are listed in Table 4. For the colors, the values are hundredths (lo-‘) of a foot-lambert of monochromatic light added to the matching center to produce a hue shift equal to that in the contrast center. For brightness, the values are the density of the wedge in the beam of the matching center. Since the density setting to make the contrast and matching centers equal in luminance was 0.35, values greater than 0.35 indicate a contrast effect. individual subjects differed in their mean levels, as is normal in such experiments,

921

Color and brightness contrast effects

Table 1. Average contrast magnitudes Stimulus

Color 1 (11= 3)

Color II (n = 3)

Color III (II = 2)

(10-a It L)

(lo-? ft L)

(10-r ft L)

I .22

2.79 3.15 3-00 3.56 4.28 4.76 5.36 7.98 3.60 3.70 4.86 740 4.16 4.62 6.43 8.39 4.57 4.76 6.23 9.13 4.46 4.74 5.89 6.00 8.71 5.23 5-57 5.81 8.28 5.4’ 5.81 6.87 8.46 5.79 9.25 5.53 8.73 6.48 9.65 9.71

8.20 8.77 8.49 7.92 8.49 9.8 1 9.83 12.37 9.05 9.22 944 1240 9.42 944 9.61 13.38 1085 9.67 Il.39 12.80 IO.87 IO.70 9.63 11.11 12.52 10.65 10.90 11.56 13.49 IO.52 11.30 10.88 13.02 I@49 13.47 11.19 14.91 10.99 13.95 1414

AB* BC DE* FG HI* IJ JK’ KL* AC FH* GI JL* AD* DC HK IL* AE BF

I.57 1.43 164 1.45

I .56 180 3.03 1.71 1.57 I.54 3.35

1.60 2.21 2.52 3.18 1.74 2.30 230 3.47 2.13 2.14 2.17 2.36 3.20 1.72 1.68 189 3.61 2.00 1.93 2.16 3.48 2.23 3.50 1.92 3.70 2.40 380 3.59

GK* HL AF* CH* EJ FK

CL* AC CI

DJ FL AH* BI DK* EL* AI DL

AJ’ CL’ AK* BL AL*

Brightness (II = 3) (wedge density) 048

0.49 0.50 0.53 062 0.53 0.62 0.49 0.63 057 0.52 0.52 0.67

0.55 0.58 0.70 0.55 0.69 0.60

-

0.7 I

Stimulus designations and the order of the stimuli is the same as Table 1

but each subject’s relative values are essentially the same as the averages. The results are in general agreement with those of previous studies. Given a constant center size, the larger the surround the greater the contrast effect; in the stimulus sequence with the smallest centers (second letter stimulus label. L) the contrast effect increases as the surround size increases (first letter of label going from K to A). Given a constant surround outer radius, the smaller the center the greater the contrast; in the stimulus sequence with the largest surround outer radius (first letter A), the contrast increases as the center size decreases (second letter going from B to L). However, further analysis is necessary to determine the specific nature of the critical spatial aspect of the stimulus. ANALYSIS

The goal of this study was to find the spatial variable, i.e. the mathematical expression of center and surround radii, that was most closely related to the magnitude of the contrast effect in the center-sur-

round stimulus. One would like to test the spatial variables for all models of contrast effects. Unfortunately, it is not possible to take the spatial variables from the various contrast models, since most of these models are not yet specific in this respect. Four representative types of spatial variables were chosen, each for the reasons indicated briefly below. I. If color and brightness are determined solely at the edges between stimulus areas, the areas being “filled in”, then the contrast magnitude might depend on the relative positions of the two edges, i.e. the width of the surround. Such a way of determining color would be consistent with the results of Krauskopf (1963), where stabilization of the edge of a colored area not only caused the edge to disappear but also caused the entire area to take on the color of its surround. 2. Several previous studies of contrast effects as a function of spatial parameters (Diamond, 1955; Kinney, 1962; Stevens, 1967) have presented their results in terms of contrast effects increasing with surround (or inducing field) area. Although no models

E. WILLIAMYUNDand JOHNC. ARMINGTOK

922

were implied. the possibility that the contrast effect is simply a function of surround area should be considered. 3. Helson’s (1964) adaptation level theory suggests that an average reference color (the adaptation level) is calculated on the basis of stimulus areas. Then the magnitude of the contrast effect should depend upon the value of the adaptation level. Exactly how to calculate the adaptation level for a circular centersurround stimulus has not been specified, but the log of the ratio of the surround to center areas seems to be a reasonable approximation. 4. Yund’s edge-distance model suggests that the positions of both edges of the stimulus with respect to the center point are critical in determining the color at that point. Edges further away from a point are weighted less in the calculation of color and brightness. In such a system, the contrast effect for a circular center-surround stimulus would be a function of the difference in the distance weighting factors for the center and outer surround edges. (See appendix for a formal treatment of the disk-annulus stimulus.)

Each of these possible determinants of contrast magnitude can be expressed mathematically in terms of the outer surround (R,) and center (RI) radii: (1) surround width, RI-R7 ; (2) surround area. RI-R;; (3) area ratio. (RI-RI)/Ri; and (4) edge-distance expression, 1/R2-l/R,. The data analysis which follows will consist of graphical and correlational comparisons between the average contrast magnitudes and the values of these spatial variables. Before the analysis is begun. it is important to note that these spatial variables are themselves interrelated. All increase as the surround outer radius (R,) increases; all decrease as the center radius (R2) increases. If, during an experiment. only one of these radii varies, all four spatial variables are monotonic functions of that radius; one value for one variable corresponds to one and only one value of each of the others. In this case it would be impossible to conclude that the contrast effect was more closely related to one or the other spatial variable, since each is exactly related to all others. Table 5 lists the values of the spatial variables for the center-surround size combinations used in this study, where

Table 5. Values of the spatial variables for the stimuli used in these experiments Stimulus AB*

BC DE* FG HI’ IJ JK* KL* AC FH* GI JL’ AD* DG HK IL* AE BF GK* HL AF’ CH* EJ FK GL* AG CI DJ FL AH* BI DK’ EL* AI DL AJ* CL* AK* BL AL* Stimulus in

Table I.

designations

Surround width (degrees)

Surround area (deg*/N

Area ratio (surr./cent.)

Edge-distance (deg- ’ x 100)

0.8677 08722 0.8800 0.8865 0.8908 0.8925 0.8934 0.8942 1.7389 1.7750 1.7793 1.7885 2.6144 26493 2.6776 2.6810 3.4944 3.5105 35661 3.5718 4.3772 4.4133 44411 4.4526 4.4603 5.2637 5.3041 5.3211 5.3468 6.1522 6.1763 6.2154 6.2296 70430 7.1096 7.9355 7.9851 8.8298 8.8573 9.7240

17,6565 16.2519 13.3144 IO.2815 7.1652 5.5873 4.0007 2.4009 33.9084 19.0092 15.8929 6.4016 48.6916 35.3965 16.7532 11.9889 62.0060 56.1501 25.4809 19.1541 73.8066 58.9074 43.5623 357624 27.8818 84.088 1 660726 56.8767 38.1633 92.8158 82.3245 60.8774 49.9639 99.9810 63.2783 105.5683 78.06 I5 109.5690 94.3134 Ill .9699

0.1856 0.2060 0.2622 0.3584 05601 0.7756 1.2492 2.9947 0.4299 a9525 1.2425 7.9850 0.7598 I.2340 5.231 I 14.9543 1.2214 1.4410 7.9563 23.8918 1.8941 2.9518 60475 11.1666 34.7783 2.9315 5.1657 7.8959 47.6029 4.6510 6.4363 19.0087 62.3224 7.8167 78.9301 14.6555 97.3699 34.2125 1176417 139.6655

0.8377

are the same

as Table

1. Stimuli

are listed

in the order

of increasing

lWJ70 1.5428 2.6516 5.5756 9.2981 18.6004 55.8034 1.8438 6.3653 9.2893 74.4227 3.0754 6.1794 33.4930 83.7207 4.6183 5.7664 37.2066 89.2965 6.6032 11.1247 23.2241 39.8584 93.0101 9.2549 16.7004 24.7670 95.6618 12.9686 17.7074 43.3861 976467 18.5443 99.1896 27.8424 100.4212 46.46 16 101.4283 102.2651 surround

width,

as

Color

and

brightness

both radii were varied. Although care was taken to include size combinations where the value of one variable remained constant and the other varied over as large a range as possible, there were still high correlations between variables. For the 40 sizes used here. the product moment correlation coefficient is 0.82 between surround width and log area ratio and 088 between log area ratio and edge-distance expression. Thus, if the magnitude of the contrast effect were an exact linear function of surround width, it would also show a correlation coefficient of 0.82 with the log area ratio. Similarly, if the contrast effect were exactly related to the log area ratio, it would correlate 088 to the edge-distance expression. This problem must be considered further in the analysis of the data for the individual colors. The analysis proper will begin with the data for color combination I. Since the analysis for each of the other colors and the one brightness level tested agrees well with that for the color I, only color I will be discussed completely. The discussion of the analyses for colors II and III and for brightness contrast will briefly illustrate their agreement with that for color I. Color combination I

The initial graphical analysis for color combination I is presented in Fig. 3. Here the average contrast magnitude is plotted on the ordinate against each of the four spatial variables on the abscissae. The

contrast

923

effects

magnitude of the contrast effect is expressed, as in Table 4. in hundredths of a foot-lambert of monochromatic light added to the matching center. For the purpose of the analysis it is best to think of the ordinates as arbitrary units of contrast magnitude. Least-squares best-fit lines are included. The top left graph of Fig. 3 illustrates that the magnitude of the contrast effect is not closely related to the surround width. There is considerable scatter about the line (S.E. 0.67) with both high and low values of average contrast occurring throughtout the range of surround widths. The situation is similar for surround area, where the S.E. is 077. The average contrast seems to be more closely related to the other two spatial variables. The scatter about the line for log area ratio (lower left) is reduced (S.E. 0.41). but it is still considerable, especially in the range of area ratios from 1 to 20. In the lower right graph of Fig. 3, the contrast magnitude seems to be most closely related to the edge-distance expression. The scatter about the line is small (S.E. D23) throughout the range of the expression. The difference between log area ratio and the edge-distance expression can be further illustrated by considering the relationship between them. The hvo lower graphs of Fig. 3 are replotted on the left in Fig. 4. The different symbols, to indicate the center size, help to emphasize the way in which the points for different stimulus sizes deviate from the least-squares line. Note that a pattern becomes apparent in the top left, log area ratio plot. All the 4.00,

l

loo

0

20

40 60 60 Surround width

I)0

0

”

20

40

Surround

0

20

40

-.

60

83

Km

120

oreo

60

80

!oo

Edge - distance

Fig. 3. Initial graphical 1, the blue-green center Constrast is plotted on in the matching center

analysis for color I. Average color contrast magnitudes for color combination and blue surround. are compared graphically with the four spatial variables. the ordinate as the luminance of monochromatic light (in 10T2 ft-L) needed to equal the effect of the surround on the contrast center. The four spatial variables on the abscissae are in the units listed in Table 5. Each of the graphs includes the least-squares best-fit line.

924

E.

Areo

WILLIAM YUND

and

rot10

Edge - dlstonce

JOHN C. AHMIWSTOS

Areo

rat10

Edge - dstonce

Fig. 4. Further analysis for color combination I. The left graphs here are those on the bottom of Fig. 3 replotted with different symbols to indicate center size. The five smallest centers in order of increasing size are represented by erect triangles, diamonds, hexagons. inverted triangles and squares. In the top right. the edge-distance predicted contrast read from the least-squares line in the lower left, is plotted as a function of the log area ratio. In the bottom right, the log area ratio predicted contrast, read from the least-squares line in the upper left, is plotted as a function of the edge-distance expression. Note the similar patterns of symbols in the top graphs and the lack of similar patterns in the bottom graphs.

points for stimuli with the smallest centers (erect triangles) lie above the line and deviate further from the line as area ratio decreases. Points for the next three larger centers (diamonds, hexagons and inverted triangles) tend to fall below the line and deviate more as area ratio increases. Such a regular pattern of “variability” indicates that some other spatial variable might be controlling the contrast magnitude. If so, this controlling variable should be able to predict the pattern. The top right graph shows the pattern of variability predicted by the edge-distance expression. The edge-distance predicted contrast (ordinate) was obtained from the equation for the least squares line of the lower left graph, where average contrast magnitude is plotted as a function of the edge-distance expression. For each center-surround size combination used, the value of the edge-distance expression was substituted into the equation for the line and a predicted contrast magnitude obtained. Then these predicted values were plotted as a function of the area ratio to obtain the upper right graph. A comparison of the upper two graphs of Fig. 4 demonstrates that the edge-distance expression does make a reasonable prediction of the pattern of the variability found for the log area ratio. The upper left graph looks like the upper right graph plus some random variation. A similar pattern of variability is not present when the average contrast magnitude is plotted

as a function of the edge-distance expression, lower left of Fig. 4. The lower right graph represents the pattern which should be present in the lower left if contrast were determined by the log area ratio. but there is no indication of such a pattern. The above graphical analysis shows that the magnitude of the contrast effect for color combination I is more closely related to the edge-distance expression than to any of the other three spatial variables. A statistical analysis of the correlation coefficients between the contrast magnitude and the spatial variables is a further test of these relationships. Table 6 gives the product moment correlation coefficients of each of the four spatial variables with the average contrast effect and with contrast predictions for log area ratio and the edge-distance expression. The predictions are read from the regression lines as in the above graphical analysis. From this table, it is clear that the edge-distance spatial variable correlates best with the average contrast. This correlation coefficient. 0.9549, is significantly greater than the next best value, O+NO, for log area ratio, at the @Ol level. (All probability levels were determined by the Z’-test for the difference between correlation coefficients.) In addition, the edge-distance pseudo-observer is an excellent predictor of the correlation between average contrast and the other three spatial variables. In each of the three

915

Color and brightness contrast effects Table 6. The product moment correlation coefficients of the various spatial variables with the average contrast color combination I and with the log area ratio and edge-distance predicted contrast Surround width

Surround area

Log area ratio

Edge-distance

0.51” 0.8236 0.5253

01913 0.520 01630

5~8500 0.8795

0,954s 0.8795 -

Color contrast I LAR prediction ED prediction

cases. &hecorrelation of the spatial variable with average contrast is not si~n~~n~y different from its correlation with the edge-distance (ED) predictions (05122 vs 05253. O-1913 vs 0.1630 and O&500 vs 0.8795; no significant differences). The situation is different for the log area ratio (LAR) predictions; the correlation of two of the three spatial variables with average contrast is significantly different from its correlation with LAR predictions (05122 vs M236, P c 005; 0.1851 vs 05179, ns., 0.9548 vs 0.8795, P < 0.05). Color combinations II and III The initiaf graphical analyses for coior combinations II and III are not presented because they dosely resemble that for color combination I in Fig. 3. In neither case was the magnitude of the contrast effect directly related to surround width or surround area. The relationships among the contrast magnitude, the area ratio and the edge-distance expression do deserve further consideration and are illustrated in Figs. 5 and 6 for color combiMtions II and III respectively. These graphs are similar to those of Fig. 4 for color combination I and again use different

symbols to indicate center size to emphasize patterns of deviation from the least squares line. As in Fig. 4, a pattern of “variabiiity” from the line is apparent in the log area ratio graphs (upper left, Figs. 5 and 6). but not in the edge-distance graphs (lower left, Figs. 5 and 6). Furthermore, the log area ratio pattern is that predicted by the edgedistance expression (compare the upper graphs of Figs. 5 and 6, as in Fig. 4). Thus, the analysis for colors II and III compares well with that of color I. The correlation coefficients for colors II and III are given in Table 7. The top two rows are the correlations between average contrast magnitudes and the four spatial variables. The bottom two rows are the correlations between the LAR and ED predictions and the spatial variables as in Table 6. For both colors, average contrast correlates best with the edgedistance expression, but in neither case is there a sign&ant difference ‘between edge-distance and log area ratio correlations. In both cases. edge-distance and log area ratio correfations with average contrast are si~i~~n~y higher than those for surround width and surround area (P c 0001). As in the case of coior combination I, there are no sign&ant differences

so tq 6-O. 3

7.0.

$60. ho. D I 4.0(

20' ,o-'2

Area

:,$Z Area

ratio

5 ol 2 .'i">,

ratio

IOOr

Edge

-

d&once

for

Edge

-distance

Fig. 5. Graphical analysis for color II. Same as Fig. 4, except that the contrast data is for color combination 11, the yellow-green center in a green surround.

926

E.

WILLY~MYti~a and

JOHN C. ARMINGTON

7oO+

Edge

Edge- distance

- distance

Fig. 6. Graphical analysis for color III. Same as Fig. 4, except that the contrast data is for color combination III, the orange center in a red surround. between the other spatial variables’ correlations with average contrast and their correlation with the ED predictions. However, the correlations for color combination II with surround width and with the edgedistance expression are signifi~n~y different from the LAR prediction’s correlations to these variables (P < 0.05). Brightness

An initial graphical analysis for brightness, not presented here, would look like that for color I in Fig. 3. The magnitude of the brightness contrast effect is not closely related to surround width or surround area. The relationship between brightness contrast magnitude and the other two spatial variables, log area ratio and edge-distance expression, is illustrated in Fig. 7, again using different symbols to indicate center size. As with color contrast, brigh~ess contrast (ordinate, left graphs) is more closeiy related to the edge-distance expression (lower left) than to the log area ratio (upper left) and the pattern in the log area ratio graph is predicted by the edge-distance expression (compare upper graphs of Fig. 7).

The correlation coefficient for the brightness contrast magnitude and the spatial variables are given in the top row of Table 8. The edge-distance expression correlation is sign~~ntiy higher than that for surround width {P < 04001}. surround area (P < O-0001) and log area ratio (P < 005). None of the other variables’ correlation with contrast is significantly higher than its correlation with the edge-distance prediction. However, the edge-distance correlation with contrast is higher than its correlation with the log area ratio prediction (P < OQOS). DlSCUSSION

The above analysis indicates that the magnitude of the color or brightness contrast effect in the center of a circular center-surround stimulus is closely related to the magnitude of the edge-distance expression for that stimulus. In fact, the correlation coe& cients for contrast magnitude and the edge-distance expression are so high that they seem to be limited by the experimental variability in the determinations of contrast magnitude. It is difficult to test this possi-

Table 7. The product moment correlation coefficients of the various spatial variables with the average contrast color combinations II and III and with the log area ratio and edge-distance predictions

Color contrast II Color contrast III LAR prediction ED prediction

Surround width

Surround area

Log area ratio

Edge-distance

0.596 1 0.6545 08236 05253

0.258 1 0.3752 0.5220 @1630

0.9235 0~9019 0.8795

0.9574 0.9087 0.8795 -

for

Color and brightness contrast effects

04001 0

0

20406080100 Edge

20

927

40

60

80

!al

Edge -distance

- distance

Fig. 7. Graphical analysis for brightness. game as Fig. 4, except that the contrast data is from the brightness contrast experiment.

Table 8. Product moment correlation coefficients of the various spatial variables with the average brightness contrast effect and with the log area ratio and edge-distance predictions

Brightness contrast LAR prediction

ED prediction

Surround width

Surround area

Log area ratio

Edge-distance

05656 0.8236

0.2617 05220

0.9074 -

0.9785 0.8795

O-5253

0.1630

08795

-

bility formally, but it is supported by the result that color I and II contrast magnitudes have higher corre-

distance or of the position of edges: ehes closer to a point are weighted more heavily in determining the

lations with the edge-distance expression (0.95 and O-96, respectively) than they have with each other (O-94). If some other spatial factor played a significant role in determining the magnitude of the contrast effect, the correlation between the results for the two colors should be greater than their individual correlations with the edge-distance expression. The differences found are in the opposite direction and they are not statistically significant; they should not be, if they are to be attributed to experimental variability. In any case, there is no indication that any factor, other than the edge-distance expression, is needed to explain the variation of contrast magnitude with spatial parameters. The edge-distance model (discussed more thoroughly in Yund, in preparation) has two basic assumptions. The first concerns the importance of edges: the model assumes that the visual system uses

color and brightness at that point than edges that are firrher away. Neither the idea that edges play a criti-

only the information concentrated at the edges of the stimulus to determine the perceived color and brightness. The second assumption concerns the role of

cal role in color nor the idea that distance-weighted interactions produce contrast effects is new with this model; the virtue of the model is its successful integration of these two ideas resulting in the quantitative prediction of the contrast effects actually found. A number of psychophysical results have indicated the importance of edges to perceived color and brightness. Of particular interest are the experiments of O’Brien (1958) and of Krauskopf (1963). Both of these experimenters used circular center-surround stimuli to demonstrate that the edge between the center and surround was a critical determinant of the brightness (O’Brien) and of the color (Krauskopl) of the center. In O’Brien’s experiments, both an edge and a brightness difference between the center and surround were perceived when the areas differed in luminance and had a sharp edge between them. However, when this sharp luminance transition was replaced by a gradual one such that an edge was no

93

E. WILLIAM YUND and JOHN C. ARMINGTOK

longer apparent. the brightness difference also disappeared. It was even possible to combine a large gra-

dual transition with a smaller sharp transition of opposite polaritt to produce a stimulus in which the center looked brighter than the surround even though the surround had the higher physical luminance. Krauskopf studied color using the stabilized image technique. When the edge between a red center and its green surround was stabilized, the entire area of the stimulus took on the color of the surround-appearing as a large uniform green disk. It should be clear that the results of O’Brien and of Krauskopf can be explained in terms of the edgedistance model because of its first assumption. If O’Brien’s gradual luminance transitions were not sharp enough to be detected as edges, then, according to the model thes should not have contributed to the subject’s perception of brightness. The brightness would be determined by those edges which were detected. In Krauskopfs experiment. the center-surround edge disappeared because it was stabilized. Once that edge could not be detected, there would be nothing to indicate that the color of the center was different from that of the surround. The color of the whole stimulus area would be determined by the still unstabilized outer edge of the surround. Thus, the results of O’Brien (1958) and of Krauskopf (1963) support the first assumption of the edge-distance model. Psychophysical tests of the model’s second assumption have been the primary topic of this report. These results demonstrate that a distance-weighted summation of edge magnitudes adequately accounts for the measured variation in center color and brightness as a function of center and surround size, and that it does so far better than other suggested factors, such as surround area and area ratio. The edge-distance model also receives support from much recent nemophysiological data. In primates, subcortical explanations for color and brightness contrast seem to be ruled out by neurophysiological evidence. No color contrast effects have been found in the responses of primate lateral geniculate cells (Wiese1 and Hubei. 1966; De Valois and Pease. 1971). De Valois and Pease (1971) measured the “brightness contrast effects” which do occur at this level and round them to cover distances too short to account for brightness contrast effects seen psychophysically. De Valois and Pease (1971) have therefore concluded that neither brightness nor color contrast effects can be explained by the responses of lateral geniculate (or retinal) cells; “some other presumably cortical process” is necessary to deaf with these effects. The edge-distance mode1 is a “cortical” model for color and brightness contrast effects. The assumptions of the model could be viewed as logical abstractions of simple and complex cortical cell types reported by Hubel and Wiesel (1962, 1968). In any case, ceils which respond optimally to single edges and other cells which seemed to be summing the responses of those cells have been found. Zeki (1973 and personal communications) has recorded from cells in prestriate cortex which might be carrying out a distance weighted summation of edges within an opponentcolor system. More quanti~tive results are necessary before precise comparisons can be made. It should be noted, however. that the distance weighting func-

tion which must be found in these cells. if they are to serve as an explanation for color and brightness contrast effects. has been defined by the work presented in this paper. The distance weighting function needed for the model is an inverse proportionality to distance. Quantitative data have become available for one complex edge detector from the macque monkey (De Valois. Hepler and Yund. unpublished observation). This cell was in cortex buried beneath striate cortex (the actual target area of that study) that was probably not striate itself. The cell’s response to a stationary single luminance edge in various positions across the field is shown in Fig. 8. plotted symbols and dashed line. The luminance difference across the edge was 1 log unit. The ordinate represents the average increase in the cell’s response over the background rate for four l-see presentations of the edge. The abscissa represents the relative positions of the edge. As the position of the edge changes from left to right. the cell’s response first rises sharply to a peak. It is as if the edge has come into the cell’s receptive field. As the edge is moved further to the right, the response falls off rapidly but not down to the background rate. The edge is not leaving the receptive field, but merely causing Less and less of a response in the cell as it moves further

and further to the right. This type of response function would be expected if the complex cell were computing a weighted sum of simple cell outputs. And since the stimulus is the same single edge each time in a different position the form of this response function should reflect the form of the distance (or position) weighting function. The distance weighting function used in the edge distance model (l/D] is pfotted in Fig. 8 as a solid line. A comparison of the two curves of Fig. 8 indicates that, in the range of S’-80’ (0.130-1.3”) and within the experimental error of such neurophysiological measurements, the distance weighting function of this cell is the same as that of the edge-distance model. Larger distances remain to be tested and the range between 0’ and 8’ would

_

. -4b-G-24 -i6-i i, e k i ai 40 i6 i6 64 R

Positron.

60

mm of vlsuaiongla

Fig. 8. A comparison of the distance weighting factor found in a primate complex edge detector ( x -- x ) with that of the level two edge detectors of the edge-distance model (solid line). The ordinate represents the average response of the cell, in spikes/set, to four I-set presentations of the stimulus, a 1 log unit black-white edge. The abscissa represents the relative position of the edge. The location of the 0 point on the abscissa is determined by the model’s distance weighting function, l/D. The cell’s response curve was adjusted laterally for the best fit to the solid tine.

929

Color and brightness contrast effects have to be explored in greater &tail i~luding a consided ration of the model’s resolution factor (K, in the Appendix). in any case, the agreement between this neurophysiological result and the edge-distance model, is quite encouraging. Acknowledgements--We are indebted to L. Marie Behenna (Blooms Peter Kason and Mary Alice Yund who participated as subjects. L. Marie Behenna (Bloom) also assisted in running the experiments and in analyzing the data.

REFERENCES

Bektsy G. von (1968) Mach- and He&g-type lateral inhibition in vision. &ion Res. 8, 1483-1499. De Valois R. L. and Pease P. L. (1971) Contours and contrast: responses of monkey Iateral geniculate nucleus cells to luminance and color figures. Science, N.Z 171, 694-696. Diamond A. L. (1955) Fovea1 simultaneous contrast as a function of inducing-field area. J. exp. Psychof. SO, 144-152. Dunn B. and Leibowitz H. (1961) The effect of separation between test and inducing fields on brightness constancy. J. exp. Psychoi. 61, 505-507.

Heinemann E. G. (1972) Simultaneous brightness induction. In: Vfsua~ Psychophysics (Edited by Jameson D. and Hurvich, L. M.). Springer-Verlag, New York. Nelson H. (1964) Adapration-Level Theory. Harper & Row, New York. Hubel D. H. and Wiesel T. N. (1962) Receptive iieids, binocular interaction and functional architecture of the cat’s visual cortex. J. Physioi., Land. 160, 108-154. Hubel D. H. and Wiesel T. N. (1968) Receptive fields and functional architecture of the monkey striate cortex. J. Pftysiol., fond. 195, 215-243. Jameson D. and Hurvich L. M. (1961) Opponent chromatic induction: ex~riment~ evaluation and theoretical account. J. opr. Sot. Am. 51, 46-53. Kinney J. A. S. (1962) Factors affecting induced color. Vision Res. 2, 503-525. Krauskopf J. (1963) Effect of retinal image stabilization on the appearance of heterochromatic targets. J. opr. Sot. Am. 53, 741-744. Leibowitz H., Mote F. A. and Thurlow W. R. (1953) Simultaneous contrast as a function of separation between test and inducing fields. .I. exp. Psychol. 46, 453-456. Mackavey W. R., Bartley S. H. and Casella C. (1961) Measurement of simultaneous brightness contrast across the retina. J. Psychol. 52, 241-250. Marsden A. M. (1969) An elemental theory of induction. Msion Res. 9, 6.53-664. O’Brien V. (1958) Contour perception, illusion and reality. J. opt. Sot. Am. 48, 112-119. Oyama 1. and Hsia Y. (1966) Compensatory hue shift in simultaneous color contrast as a function of separation between inducing and test fields J. exp. Psychol. 71,405-413. Stevens J. C. (1967) Brightness inhibition re size of surround. Percept. Psychophys. 2, 189-192. Wallach H. 11948) Bri~htne~ constancv and the nature of achromatic cblors~ J. exp. Psychof. *3g, 310-324. Wiesel T. N. and Hubel D. H. (1966) Spatial and chromatic interactions in the lateral geniculate body of the rhesus monkey. J. Neurophysiol. 29, 1115-l 156. Zeki S. M. (1973) Color coding in rhesus monkey prestriate cortex. Brain Res. 53, 422-427.

APPENDIX Derizbon

of the edge-distance expression The color (or brightness) at a point within one opponent color system is given by the equation

J*=a

i=i

d@ 2x where @ = the direction away from the point in question; Ei = the magnitude of the edge within this color system; Di = the distance between the point and the edge measured along a line at direction Q,from the point; 0 = the deviation of the edge from perpendicular to that line: and Ki = a resolution factor which prevents this function from appr~ching infinity as Di approaches zero-this represents a high spatial frequency falIoff in MTF terms. For the center point of a circular center-surround stimulus the equation can be simplified considerably: 1. All parameters are independent of (P, then C=

2. The edge of a circle is perpendicular to any radii1 tine, i.e. (3 = 0 in all cases: letf(0) 2’ 1, then

3. If Dt are relatively large, we can ignore K,:

For the disk-annulus

stimulus with only two edges:

++g 2

‘

where i refers to outer surround edge and 2 to the center edge. If we further assume linearity within edge detectors: E2 = Xc - Xs E, = Xs - XB

where Xc = value for center color; Xs = value for surround color; Xa = value for background color; (X, = 0 in this experiment). Then,

c

xc- xs xs =-+qDZ

where R, is the outer surround radius and R, is the radius of the center. If there were no light in the surround and therefore no contrast effect, X, would be 0 and C = Xc/ R2. This means that the contrast effect is represented by the terms X, (l/R2 - l/R,) and that it should be a linear function of (l/R? - l/R i), the edge-distance expression.