Brightness induction: Unequal spatial integration with increments and decrements

Visual Neuroscience (2004), 21, 353 357. Printed in the USA. Copyright 2004 Cambridge University Press 0952-5238004 $16.00 DOI: 10.10170S0952523804213037 Brightness induction: Unequal spatial integration with increments and decrements SANG WOOK HONG and STEVEN K. SHEVELL Departments of Psychology and Ophthalmology and Visual Science, University of Chicago, Chicago (Received September 7, 2003; Accepted January 15, 2004) Abstract Modern theories of brightness induction include an influence from regions that do not share a border with the target. This study tested whether the spatial range of neural integration is the same with incremental versus decremental contrast edges in relatively remote parts of the background. Using an asymmetric matching task, observers set the brightness of a comparison ring, within its own uniform surround, to match the brightness of a test ring within a contiguous surround and a noncontiguous background. The measurements showed that the area of integration depended on the incremental versus decremental contrast polarity at the edge between the surround and background. This implies that brightness induction from an inhomogeneous background must consider the polarity of contrast edges within the whole scene. Keywords: Brightness induction, Brightness contrast, Increments, Decrements, Contrast polarity, Edge integration Introduction A gray patch on a dim background appears brighter than a physically identical patch on an intense background. This phenomenon is called brightness induction. A classical account of induction from a homogeneous background holds that the physical contrast between the target and its contiguous surround determines brightness (Wallach, 1948). Induction from nonuniform backgrounds, however, cannot be explained by only the contrast between the target and its contiguous surround. The brightness of a target within an inhomogeneous background depends on neural signals from throughout the area of the visual stimulus (Reid & Shapley, 1988; Shevell et al., 1992; Zaidi et al., 1992; Blakeslee & McCourt, 1997; Rudd & Arrington, 2001). Some theoretical models incorporate multiscale neural filters whose outputs are scaled by contrast normalization (e.g. Blakeslee & McCourt, 1997, 1999). Other models extend Wallach s theory by positing integration of neural signals from each luminance edge in the scene, with the strength of influence of an edge decreasing as the distance from the target increases. The present study investigated how the spatial range of neural integration depended on the polarity of the contrast edge between the contiguous surround and a noncontiguous background. The measurements showed that the range of spatial integration depended on the Address correspondence and reprint requests to: Steven K. Shevell, Visual Science Laboratories, The University of Chicago, 940 East 57th Street, Chicago, IL 60637, USA. E-mail: shevell@uchicago.edu contrast polarity. While these results are most directly related to edge integration models (e.g. Rudd & Arrington, 2001), they also reveal a new property of brightness induction that should be implicit in any account of brightness perception. Materials and methods Apparatus Achromatic stimuli were generated using a Macintosh G4 computer, and were presented on a calibrated Sony color display (GDM-F520). The cathode ray tube (CRT) display had 1360 1024 pixel resolution and a refresh rate of 75 Hz noninterlaced. The achromatic stimuli were approximately metameric to equal energy white with Judd (1951) chromaticity coordinates x 0.33, y 0.33. The red, green, and blue guns of the color CRT were linearized using 10-bit lookup tables. Stimuli The CRT simultaneously displayed the (left) comparison field and the (right) test field side-by-side, separated horizontally by a 2.3-deg dark region (Fig. 1). The width of both the test ring and the comparison ring was 8.6 min of arc. The left comparison ring was surrounded by a uniform field at a fixed luminance of 24 cd0m 2 (inner0outer diameter of 1.4204.0 deg). The test ring, at 16 cd0m 2, was surrounded by contiguous concentric rings on either side, which in turn were surrounded by the background field (outer 353

354 S.W. Hong and S.K. Shevell Fig. 1. A schematic of the stimuli for the matching task (at left) and the luminance profile of the test field (at right). The right-side test-ring luminance was fixed at 16 cd0m 2. The test s contiguous rings were varied in luminance (18, 20, 26, and 30 cd0m 2 ) as indicated by the double-pointed arrows in the luminance profile, and in width (4.3 34.4 min of arc). The left-side comparison-ring luminance was adjusted by the observer to match the appearance of the test ring. The comparison-ring surround luminance was fixed at 24 cd0m 2. diameter of 4.0 deg). The contiguous rings were varied in luminance (18 30 cd0m 2 ) and width (extending between 4.3 and 34.4 min of arc either side of the test ring). The noncontiguous background was fixed at the same luminance as the comparison background (24 cd0m 2 ) in most experiments, and was varied in size to fill the part of the test field not occupied by the test and its contiguous rings. In Experiment 2, an additional condition was tested with the contiguous rings at 26 cd0m 2 and the noncontiguous background at 30 cd0m 2. Procedure Observers participated in several practice sessions until they made stable matches. A chin rest was used to maintain a stable head position. In each session, between six and 12 conditions were run in a random order. When a condition began, the test field was presented on the right side of the CRT and observers adapted to it for 2 min. The comparison field then appeared and observers made one match followed by 1 min of adaptation to the whole stimulus display. Observers were instructed to look at the presented stimuli during both adaptation periods. This trial match was excluded from data analysis. The average of five following matches for each condition was taken as one measurement for further data analysis. Each condition was repeated three times on different days. The method of adjustment was used to set the luminance of the comparison ring to match the brightness of the test ring. Observers controlled the luminance of the comparison ring by pressing separate buttons on a joystick. Observers were free to view the stimuli during each trial (no steady fixation). The initial value of the comparison ring in each trial was randomized. Observers Three observers participated in the study. They had normal color vision as tested with a Neitz anomaloscope. Author S.W.H. was knowledgeable about the experimental design and had prior experience making brightness matches. Observers Y.I.O. and L.Y. were naïve. Consent forms were completed in accordance with the policy of the University of Chicago s Institutional Review Board. Results Experiment 1: Spatial integration with remote increments and decrements The brightness matches showed that the influence of the noncontiguous background decreased as the distance between the edge of the background and the test increased, which is consistent with published brightness models (Reid & Shapley, 1988; Rudd & Arrington, 2001). The measurements, however, also showed a difference in the range of spatial integration for incremental and decremental edges. The matches are plotted in Fig. 2, in which the abscissa is the width of each contiguous ring, and the ordinate is the observer s setting of the comparison-ring luminance to match the brightness of the test ring. Recall that the test-ring luminance was fixed at 16 cd0m 2, and both backgrounds were 24 cd0m 2. The luminance of the contiguous rings is indicated by symbol shape (circles, 18 cd0m 2 ; squares, 20 cd0m 2 ; diamonds, 26 cd0m 2 ; and triangles, 30 cd0m 2 ). Each gray horizontal band shows the two-standarderror range of the measurement for which the contiguous-surround luminance filled the entire area of the test background (taken from each rightmost plotted point). The light gray band, therefore, shows a statistical range for the match to the brightness of the test ring with a single uniform surround. The smallest contiguous-ring width for which the brightness match was within the gray band was taken as an estimate of the extent of spatial integration, because further increasing contiguous-ring width did not significantly change the brightness of the test. Open arrows show this ring width for conditions in which the luminance of the contiguous rings was lower than the luminance of the noncontiguous background (pure decremental condition, test contiguous surround background). Black arrows show this ring width for the conditions in which the luminance of the contiguous rings was higher than the luminance of the noncontiguous background (test contiguous surround background). When the luminance of the contiguous rings was lower than the luminance of the noncontiguous background (circles and squares, Fig. 2), the brightness of the test ring increased as the width of the contiguous ring increased until it reached 26 30 min of arc for

Spatial integration with increments and decrements 355 brightness models that include neural integration from each luminance edge in a visual scene, with a decrease in the strength of the induction as the distance from the target to the edge increased (Reid & Shapley, 1988; Rudd & Arrington, 2001). When the luminance of the contiguous rings was higher than the luminance of the noncontiguous background (diamonds and triangles, Fig. 2), so that the contrast polarity of the noncontiguous edge (increment) was different from that of the contiguous edge (decrement), the brightness of the test ring reached asymptote more abruptly than for the pure decremental condition. The brightness of the test ring reached asymptote when the width of the contiguous ring was 4.3 8.6 min of arc (compare to 26 30 min of arc) for observer S.W.H., 4.3 17.2 min of arc (compare to 26 min of arc) for Y.I.O., and 4.3 min of arc (compare to 34 min of arc) for observer L.Y. In sum, when the contrast polarity between the contiguous rings and background was different than the polarity between the test and its contiguous rings, the brightness of the test was affected by inducing light over a shorter distance (compare black to open arrows in Fig. 2). Fig. 2. Brightness matches as a function of the width of the contiguous rings (horizontal axis). The vertical axis is the observer s luminance setting for the comparison ring to match the brightness of the test ring (fixed at 16 cd0m 2 ). The symbol shape indicates the luminance of the contiguous rings (circles, 18 cd0m 2 ; squares, 20 cd0m 2 ; diamonds, 26 cd0m 2 ; and triangles, 30 cd0m 2 ). The luminance of the noncontiguous background was fixed at 24 cd0m 2. Error bars, from three different days measurements, were usually smaller than the symbol size. Each gray horizontal band represents the two-standard-error range of the measurement when the contiguous rings filled the entire area of the test background (rightmost points). Arrows at the bottom of each panel are the widths at which the brightness of the test ring reached asymptote (see text). Each panel shows results for a different observer. Experiment 2: Remote-edge contrast polarity, not magnitude of contiguous contrast A smaller range of spatial integration was found with edges of opposite polarity but perhaps this range reflected higher contiguousedge contrast, due to increasing the luminance of the contiguous surround, rather than the difference in contrast polarity between the contiguous and noncontiguous edges. Measurements with an additional luminance condition tested this possibility. In this experiment, the luminance of the contiguous rings was 26 cd0m 2 and the luminance of the background was 30 cd0m 2. The contrast at the contiguous edge was the same as in one of the previous conditions (diamonds, Fig. 2) but the contrast polarity of the noncontiguous edge was now a decrement. In this experiment, therefore, the contrast polarity between the contiguous rings and background was the same as between the test and the contiguous rings. If the magnitude of contrast at the contiguous edge determined the range of spatial integration, then the brightness of the test ring should asymptote abruptly, as was found for the 26 cd0m 2 surround in the first experiment (diamonds and leftmost black arrows, Fig. 2). Measurements for two of the observers showed instead long-range spatial integration: 26 min of arc for observer S.W.H. and 22 min of arc for observer L.Y. (open arrows, Fig. 3). These distances are comparable to the pure decremental case in Experiment 1 (circles and squares, Fig. 2), and substantially larger than the spatial integration found with the 26 cd0m 2 surround in Experiment 1 (26 vs. 4.3 min of arc for S.W.H., 22 vs. 4.3 min of arc for L.Y.). For observer Y.I.O., the range of spatial integration by our two-standard-error criterion was smaller than for the other observers (8.6 min of arc), but still larger than in Experiment 1 for this observer (4.3 min of arc with the 26 cd0m 2 surround). The short range for Y.I.O. in Experiment 2 resulted primarily from the large standard error with the uniform surround (note the large width of the gray band for this observer in Fig. 3). In this case, the estimate of the region of spatial integration was not reliable. observer S.W.H., 26 min of arc for observer Y.I.O., and 34 min of arc for observer L.Y. In each of these luminance conditions, therefore, the brightness of the test was affected by inducing light up to about 0.5 deg away. These results were consistent with Discussion This study confirmed that the brightness of a test was affected by inducing light spatially separated from the test area. The effective range of spatial integration, however, depended on the contrast

356 S.W. Hong and S.K. Shevell Fig. 3. Brightness matches as a function of the width of 26 cd0m 2 contiguous rings. The test-ring luminance was 16 cd0m 2. The luminance of the noncontiguous background was 30 cd0m 2. The ordinate, error bars, gray horizontal bands, and arrows are as in Fig. 2. Each panel shows results for a different observer. polarity at the noncontiguous edges. This dependence cannot be explained by brightness induction models that depend on a single center-surround receptive field, or by edge-integration models that do not take account of the polarity of contrast. A center-surround receptive field is often used to explain brightness induction because of its antagonistic spatial organization (Jameson & Hurvich, 1961; Kingdom & Moulden, 1989; Moulden & Kingdom, 1990). Receptive-field organization can account for the changes in brightness of a test that result from increasing the luminance of the contiguous surround, and for some brightness changes here as the size of the surround increases and thus replaces background light of a different luminance. The asymptotic brightness of the test as surround width increases would reflect a surround size that completely fills the receptive field. A center-surround receptive field, however, cannot explain the difference in the range of spatial integration between the pure decremental condition (test contiguous surround background) and the mixed polarity condition (test contiguous surround background). The contrast polarity at the noncontiguous edge should not change spatial integration. Edge-integration models posit spatial integration of neural signals representing the contiguous edge between test and surround, and the noncontiguous edge between surround and background. Rudd and Arrington (2001) propose that integration of neural signals depends on the strength of signals from both the noncontiguous and contiguous edges. They report that neural signals from a noncontiguous edge can be partially blocked by the contiguous edge, and that blocking is more complete with higher contrast at the contiguous edge. This might explain the shorter range of spatial integration in Experiment 1 with the luminance of the contiguous surround higher than the luminance of the noncontiguous background (Fig. 2), in which case greater contrast at the contiguous edge could more fully block the signal from the noncontiguous edge. Blockage, however, is posited to occur only when the polarity of the luminance edge between the noncontiguous background and the contiguous surround is a decrement (Zemach & Rudd, 2002). They report that the influence from an incremental noncontiguous edge is not blocked. Therefore, the short range of spatial integration in the mixed polarity condition cannot be explained by blocking. The results here showed that the magnitude of contrast at the contiguous edge could not explain the range of spatial integration: with contrast at the contiguous edge held constant (26 cd0m 2 contiguous surround, 16 cd0m 2 test), the extent of spatial integration depended on the polarity of contrast at the more distant edge between surround and background, with more widespread spatial integration in the pure decremental case [a recent abstract by Zemach & Rudd (2003) reports a similar observation]. Note also that a greater absolute magnitude of contrast at the noncontiguous edge did not increase the range over which it influenced test brightness. Noncontiguous edge contrast was higher in Experiment 1 (30 cd0m 2 contiguous surround, 24 cd0m 2 background) than in Experiment 2 (26 cd0m 2 contiguous surround, 30 cd0m 2 background), yet spatial integration was greater in Experiment 2. In sum, brightness induction from an inhomogeneous background depends on the polarity of luminance edges in the visual scene. When the polarity of the noncontiguous edge and the contiguous edge are both decremental, the influence of the noncontiguous background varies over a distance of about 0.5 deg. When the polarity of the noncontiguous edge is an increment and the contiguous edge a decrement, however, the extent of spatial integration is substantially less. Acknowledgments This research was supported by PHS grant EY-04802. Publication was supported in part by an unrestricted grant to the Department of Ophthalmology and Visual Science from Research to Prevent Blindness.

Spatial integration with increments and decrements 357 References Blakeslee, B. & McCourt, M.E. (1997). Similar mechanisms underlie simultaneous brightness contrast and grating induction. Vision Research 37, 2849 2869. Blakeslee, B. & McCourt, M.E. (1999). A multiscale spatial filtering account of the White effect, simultaneous brightness contrast and grating induction. Vision Research 39, 4361 4377. Jameson, D. & Hurvich, L.M. (1961). Complexities of perceived brightness. Science 133, 174 179. Judd, D.B. (1951). Report of the U.S. Secretariat Committee on Colorimetry and Artificial Daylight. In Proceedings of the Twelfth Session of the CIE, Stockholm, 1.1L. Paris: Bureau Central de CIE. Kingdom, F. & Moulden, B. (1989). Corner effect in induced hue: Evidence for chromatic band-pass filter. Spatial Vision 4, 253 266. Moulden, B. & Kingdom, F. (1990). Light-dark asymmetries in the Craik Cornsweet O Brien illusion and a new model of brightness coding. Spatial Vision 5, 101 121. Reid, R.C. & Shapley, R. (1988). Brightness induction by local contrast and the spatial dependence of assimilation. Vision Research 28, 115 132. Rudd, M.E. & Arrington, K.F. (2001). Darkness filling-in: A neural model of darkness induction. Vision Research 41, 3649 3662. Shevell, S.K., Holliday, I. & Whittle, P. (1992). Two separate neural mechanisms of brightness induction. Vision Research 32, 2331 2340. Wallach, H. (1948). Brightness constancy and the nature of achromatic colors. Journal of Experimental Psychology 38, 310 324. Zaidi, Q., Yoshimi, B., Flanigan, N. & Canova, A. (1992). Lateral interactions within color mechanisms in simultaneous induced contrast. Vision Research 32, 1695 1707. Zemach, I.K. & Rudd, M.E. (2002). Blocking of achromatic color induction signals by borders of different contrast polarities [Abstract]. Journal of Vision 2, 106a. Zemach, I.K. & Rudd, M.E. (2003). Spatial decay of achromatic color induction differs for lightness and darkness induction processes [Abstract]. Journal of Vision 3, 421a.