Stereoscopic Illusory Contours—Cortical Neuron Responses and Human

Barbara Heider1, Lothar Spillmann2, and Esther Peterhans3 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/7/1018/1757651/089892902320474472.pdf by guest on 18 May 2021

Abstract & In human perception, figure–ground segregation sug- signal such illusory contours and can be selective for certain gests that stereoscopic cues are grouped over wide areas of figure–ground directions that human observers perceive at the visual field. For example, two abutting rectangles of these contours. The results suggest that these neurons equal luminance and size are seen as a uniform surface group stereoscopic cues over distances up to 88. In addition, when presented at the same depth, but appear as two we compare these results with human perception and show surfaces separated by an illusory contour and a step in that the mean stimulus parameters required by these depth when presented with different retinal disparities. neurons also induce optimal percepts of illusory contours Here, we describe neurons in the monkey that in human observers. &

INTRODUCTION representations of such contours have been identified Illusory contours are perceived in visual scenes where in the human visual cortex by means of imaging objects occlude one another and produce overlapping techniques such as positron emission tomography images on the retina. If these objects have equal (PET) and functional magnetic resonance imaging luminance, their retinal images merge because their (fMRI). Representations have been localized early in overlapping borders lack luminance contrast. The visual visual processing, in area V2 (Larsson et al., 1999; system has developed methods to recover the segrega- ffytche & Zeki, 1996; Hirsch et al., 1995), and in higher tion of figure and ground in such situations—it gen- areas such as V3A, V4v, V7, and V8 (Mendola, Dale, erates illusory contours that complete the occluding Fischl, Liu, & Tootell, 1999). At the single-cell level, borders where they lack luminance contrast. Figure 1 representations of contours as shown in Figure 1 have mimics such a situation. It induces the perception of been identified as well as contours between abutting two white rectangles that appear to occlude one line-gratings (thin, widely spaced lines, displaced by another. The upper (occluding) rectangle is perceived half a cycle; see Soriano, Spillmann, & Bach, 1996; as being bounded by an illusory contour. This contour Kanizsa, 1979). These representations were found early is generated from interposition, or occlusion cues, as in visual processing, in the cat (areas 17 and 18) (Sheth, produced by the gray objects (line, two discs) that are Sharma, Rao, & Sur, 1996; Leventhal & Zhou, 1994; perceived as being partially occluded and located Redies, Crook, & Creutzfeldt, 1986) and in the monkey between these rectangles. These cues often have the (mainly area V2) (Grosof, Shapley, & Hawken, 1993; form of line-ends, corners, and different types of junc- Peterhans & von der Heydt, 1989; von der Heydt, tion. As illustrated in Figure 1, they are asymmetrical by Peterhans, & Baumgartner, 1984; von der Heydt & nature and point towards the occluding surface (upper Peterhans, 1989). Further, the results of Nieder and rectangle in this case). They define the location of the Wagner (1999) suggest that birds (owls) also perceive occluding object, relative to a particular contour. With- illusory contours and that their includes out these cues, the two rectangles cannot be segregated; neurons sensitive to such stimuli. the image is perceived as a single, uniform surface. Illusory contours as described above can be perceived The perception of such illusory contours has been monocularly—stereopsis is not required. However, studied extensively in humans using stimulus configu- scenes that include situations of spatial occlusion usu- rations as shown in Figure 1 (Schumann, 1900; Petry & ally also provide binocular cues. These occur in the form Meyer, 1987; Kanizsa, 1979; Varin, 1971). More recently, of binocular disparity, or as unpaired image features that are visible to one eye only and not to the other (von Szily, 1921; for overviews, see Anderson & Nakayama, 1Yale University, 2University of Freiburg, 3University Hospital 1994; Anderson & Julesz, 1995). Occlusion cues as Zurich shown in Figure 1 also interact with stereoscopic cues

D 2002 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 14:7, pp. 1018–1029

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902320474472 by guest on 03 October 2021 (Nakayama, 1996; Carman & Welch, 1992; Ramachan- dran & Cavanagh, 1985; Harris & Gregory, 1973). Neuro- nal signals of bars defined by stereoscopic cues have been found recently by Bakin, Nakayama, and Gilbert (2000) using stimuli as described by von Szily (1921) (see also Ehrenstein & Gillam, 1998, Figure 6). While these stimuli induced the perception of a narrow sur- face bounded by an illusory contour on either side, we

aimed to study single contours. Therefore, we designed Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/7/1018/1757651/089892902320474472.pdf by guest on 18 May 2021 a novel stimulus that produced a single illusory contour and step in depth in which the figure–ground direction was defined by stereoscopic cues. This stimulus con- sisted of two abutting rectangles of equal luminance and size that were presented with different retinal disparities—one rectangle usually with zero disparity, Figure 2. Illusory contour defined by stereoscopic cues. The figure the other with either crossed or uncrossed disparity. illustrates the perception of an illusory contour stimulus in Figure 2 shows how human observers perceived this which the upper rectangle was presented with crossed (near) disparity and the lower rectangle with zero disparity. In this stimulus, namely, as two plane surfaces separated by a situation, the visual system uses the stereoscopic information of the step in depth and an illusory contour bounding the contrast borders (‘‘inducing borders,’’ indicated by arrows) to nearer (upper) surface. Since we aimed to determine generate the illusory contour. To study the effects of these cues the effects of the contrast borders that induced this selectively, we introduced disparity orthogonal to these borders contour (‘‘inducing borders,’’ indicated by arrows), (see Figure 3). we introduced disparity orthogonal to these borders (see Figure 3). No segregation occurred when the two rectangles were presented with the same disparity; human observers perceive at these contours. In addi- the stimulus was perceived as a single, uniform surface tion, we studied the perception of these contours in (see inset row 4 of Figure 4A). human observers and show that the mean stimulus In the following, we show that neurons in the monkey visual cortex signal such illusory contours and show similar orientation selectivity for these contours as for contours defined by luminance contrast (bars, edges). Furthermore, we show that some of these neurons are sensitive to the direction of the step in depth that

Figure 3. Examples of stimuli. (A) The upper rectangle is presented Figure 1. Illusory contour defined by occlusion cues. The figure with crossed disparity (arrows), the lower rectangle with zero induces the perception of two white rectangles that appear to overlap disparity. If the reader fuses the images of the left and right panels, one another, with the upper (occluding) rectangle bounded by an the upper rectangle appears nearer than the lower rectangle and illusory contour. This perception is induced by the interposition or bounded by an illusory contour. (B) Here, the lower rectangle is occlusion cues represented by the visible parts of the gray objects presented with zero, the upper rectangle with uncrossed disparity (line, two discs) that appear to be located in between these rectangles. (arrows). Upon fusion of the images, the reader now perceives the As shown here, these cues often have the form of line-ends and lower rectangle to be nearer than the upper rectangle and bounded corners that are asymmetrical by nature and indicate the location of by an illusory contour. (The cross marks an average position of the the occluding surface. fixation target of the monkey).

Heider, Spillmann, and Peterhans 1019

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902320474472 by guest on 03 October 2021 Figure 4. Neuronal signals of stereoscopic illusory contours. (A) Responses of a neuron of area V2 to monocular stimula- tion (‘‘left’’ and ‘‘right eye,’’ respectively), to an illusory contour stimulus (‘‘binocular’’; lower rectangle zero disparity, upper rectangle 30 min arc uncrossed disparity), and to a binocular control stimulus (both rectangles zero Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/7/1018/1757651/089892902320474472.pdf by guest on 18 May 2021 disparity). Ellipses indicate the response field as mapped with a bar stimulus. All stimuli were presented at the neuron’s preferred orientation (128) and were moved parallel to the lateral borders of the rectangles (arrows). Responses recorded in the forth sweeps of stimulus movement are shown in the left half, those recorded in the back sweep in the right half of the dot displays. Each dot represents an action potential; 32 cycles of stimulus movement were recorded for each stimulus. The figures underneath each display indicate the mean numbers of spikes per stimulus cycle (frequency, 1.5 Hz; amplitude, 1.58). (B) Quantitative classification. Axes indicate t values of a statistical comparison of the responses to the illusory contour stimulus and to the control stimulus (t1), and to the illusory contour stimulus and to monocular stimulation (t2). The dotted lines indicate critical t values (t = 1.73, p = 0.05). The diamond shows the result for neuron 7HK1 (for further explanation of the results, see text).

parameters required by these neurons also induced excluded from this analysis (for details of classification, optimal percepts of illusory contours in these observers. see Heider, Meskenaite, & Peterhans, 2000). Finally, we tested stimuli as shown in Figure 3. In these two examples, the lower rectangle is presented with zero RESULTS disparity, the upper rectangle with crossed (A) and We studied 96 neurons in the visual cortex of awake, uncrossed disparity (B). Disparity was always introduced behaving monkeys with stimuli that in human observers orthogonal to the lateral borders of these rectangles as induced the perception of stereoscopic illusory contours indicated by arrows. As a consequence, the stimuli as shown in Figure 2. Of these 96 neurons, 90 neurons included both horizontal and vertical disparity compo- were located histologically, 29 in the striate cortex (V1), nents, as explained in detail below (see Figure 6). This 59 in the prestriate cortex (53, V2; 6, V3/V3A), and two in method allowed us to study specifically the effects of the the border region between these cortices. ‘‘inducing borders’’ and keep the size of the rectangles constant at all disparities. Figure 3 further shows that the illusory contour is perceived as part of the nearer Stimulus Specifications and Mapping of Receptive (occluding) surface in these stimuli, in (A) as part of the Fields upper rectangle, in (B) as part of the lower rectangle. The stimuli were generated separately for each eye on a These stimuli are henceforth referred to as ‘‘illusory high-resolution oscilloscope and were projected to the contour stimuli.’’ animal’s eyes by means of a stereoscope (for details, see Methods). We used bars and edges to determine the Responses of Cortical Neurons neuron’s preferred stimulus, its preferred orientation, and the size and position of the receptive field. All Responses to Illusory Contour Stimuli stimuli could be moved automatically using a system Figure 4A shows the responses of a neuron of area V2 to of analog/digital circuits, or manually using a joystick. a stimulus that induced the perception of an illusory The mean eccentricity of the receptive fields studied contour as shown in Figure 3B, and to three control was 3.48 (range: 0.5–7.98, n = 90). The neuron’s conditions that failed to induce this perception. (Note preferred stimulus was also used to determine its that the insets depict percepts and not actual stimulus sensitivity to stimulus length and binocular disparity. conditions.) The responses to the illusory contour Neurons with end-stopped receptive fields were stimulus are shown in row 2 (‘‘binocular’’). The lower

1020 Journal of Cognitive Neuroscience Volume 14, Number 7

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902320474472 by guest on 03 October 2021 rectangle was presented with zero and the upper Longer stimuli usually evoked weaker responses or none rectangle with 30 min arc uncrossed (far) disparity. at all. Rows 1 and 3 show the responses to monocular stim- ulation of the left and right eye, respectively. These Disparity Sensitivity responses were much weaker than the responses to binocular stimulation (row 2). A binocular stimulus in Figure 5 illustrates the disparity sensitivity of neuron which both rectangles were presented with the same 7HK1 to a contrast border (A, edge) and to an illusory disparity, failed to evoke a response (row 4). contour stimulus (B). As stated above, we used contrast

borders to define the classical disparity sensitivity of Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/7/1018/1757651/089892902320474472.pdf by guest on 18 May 2021 cortical neurons. Figure 5A shows the result of this Classification of Neurons experiment for neuron 7HK1, which preferred uncrossed Figure 4B illustrates the quantitative criterion that we established to define the sensitivity of cortical neurons to these stimuli. Using Student’s t test, we compared the responses evoked by the illusory contour stimulus (Figure 4A, ‘‘binocular’’) with the strongest monocular responses (Figure 4A, ‘‘right eye’’ in this case) and the responses evoked by the binocular control stimulus (Figure 4A, ‘‘control’’). Neurons that gave significantly stronger responses to the illusory contour stimulus than to the control stimulus (t1 > 1.73, df = 19; p < .05) and to monocular stimulation (t2 > 1.73, df = 19; p <.05) were called ‘‘sensitive to illusory contour stimuli.’’ The plot of Figure 4B shows these t values for 44 neurons for which we were able to record complete sets of measure- ments. Based on this classification, we found 17 neurons (open dots) that were sensitive to illusory contour stimuli, and 27 (filled dots) that were not. In 19 addi- tional neurons, we recorded only one or the other control condition quantitatively, and 27 neurons were studied qualitatively. Of these neurons, we found five that gave much stronger responses to the illusory con- tour stimulus than to the control stimuli. Since this difference was clearly audible when listening to the responses, we called these neurons also sensitive to illusory contour stimuli. The proportions of neurons sensitive to illusory contour stimuli were different in the striate and prestriate cortex. In the prestriate cortex, we found 32% (19/59) of the neurons showing this sensitivity, but only 7% (2/29) in the striate cortex (x2 = 6.9, df =1,p < .01). One additional neuron was found in the border region between these cortices. This suggests that neuronal signals of stereoscopic illusory contours are common in the prestriate cortex, but rare in the striate cortex. Figure 5. Disparity sensitivity of cortical neurons. (A) Responses to disparity variations of a contrast border (edge). (B) Responses to disparity variations of the upper (filled dots) and lower rectangle (open Stimulus Length dots) of an illusory contour stimulus. Stimulus insets depict percepts In all these experiments, we carefully adjusted the induced by the disparity variation of the upper rectangle, from left to right with crossed, zero, and uncrossed disparities. General: In all length of the illusory contour stimulus (dimension along experiments, disparity was varied orthogonal to the lateral borders of the neuron’s preferred orientation) to the size of the the stimulus as indicated by arrows. Abscissa scales indicate horizontal receptive field as mapped with the neuron’s preferred disparity components—negative values for crossed (near), positive stimulus (see above). In particular, we were careful to values for uncrossed (far) disparities. Ordinate scales indicate mean place the ‘‘inducing borders’’ outside this field to avoid numbers of spikes per stimulus presentation (eight cycles of stimulus movement for each disparity: frequency 1.5 Hz, amplitude 1.58). stimulation with contrast borders within this field. We Dashed lines indicate spontaneous activity and horizontal markers always tested several stimulus lengths (range: 0.7–10.78) indicate the responses to monocular stimulation (R = right eye; and selected the length that evoked maximum response. L = left eye). For further properties of this neuron, see Figure 4.

Heider, Spillmann, and Peterhans 1021

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902320474472 by guest on 03 October 2021 (far) disparities and gave a maximum of response at tour stimuli. This result was typical for the majority of about 40 min arc. This result was similar, whether these neurons (78%, 14/18). disparity was introduced horizontally, or orthogonal to Figure 5B further shows that uncrossed disparities of the stimulus borders, as indicated by arrows. We classi- the upper rectangle (filled dots, positive disparity range) fied these responses according to the criteria of Poggio evoked stronger responses than uncrossed disparities of and Fischer (1977). Neurons that preferred zero or close the lower rectangle (open dots, positive disparity range). to zero disparities were called ‘‘tuned-excitatory’’ types, This result indicates that the neuron preferred one of neurons that preferred crossed or uncrossed disparities two possible stimulus configurations. It preferred the

‘‘near’’ or ‘‘far’’ types, and neurons with response mini- lower rectangle to be nearer (‘‘figure’’) than the upper Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/7/1018/1757651/089892902320474472.pdf by guest on 18 May 2021 ma at zero or close to zero disparity ‘‘tuned-inhibitory’’ rectangle (‘‘ground’’), and not the reverse. Since the types. Neurons that were sensitive to illusory contour illusory contour was always perceived as part of the stimuli showed the following distribution: 55% (12/22) occluding surface (‘‘figure’’) in this stimulus, we called were of the tuned-excitatory type, 41% (9/22) of the near these neurons ‘‘selective for a certain figure–ground or far type, and 4% (1/22) of the tuned-inhibitory type. direction’’ at these contours. This property was found in This distribution was similar for neurons that failed to 76% (10/13) of the neurons for which we had complete signal stereoscopic illusory contours. Of the 53 neurons sets of measurements. In the majority of these neurons, for which we had sufficient data available, 58% (31/53) this selectivity depended on absolute retinal disparity, were of the tuned-excitatory type, 34% (18/53) were of which means that the stimulus had to include the the near or far type, and 8% (4/53) were of the tuned- neuron’s preferred disparity (uncrossed in this case). inhibitory type (x2 = 0.46, df =2,p > .05). In addition, Similar configurations, but produced with nonpreferred we found that all neurons that failed to show sensitivity disparities(crossedinthiscase),failedtoevokea to binocular disparity (n = 13) also failed to signal response. Thus, the neuron failed to respond when stereoscopic illusory contours. the preferred stimulus configuration was produced by Figure 5B shows the corresponding disparity response presenting the lower rectangle with crossed (near) and functions for an illusory contour stimulus. The filled dots the upper rectangle with zero disparity. Only one neu- show the responses to disparity variations of the upper ron showed this selectivity independent of retinal dis- rectangle (lower rectangle, zero disparity), the open parity, and this neuron was recorded in V3/V3A. dots the responses to disparity variations of the lower rectangle (upper rectangle, zero disparity). The three Orientation Selectivity insets illustrate percepts of the first stimulus, from left to right with the upper rectangle presented with crossed Since orientation selectivity is a critical property of (near), zero, and uncrossed (far) disparities. A compar- cortical neurons that indicates selectivity for orientated ison of Figure 5A and B shows that neuron 7HK1 contours, it is important to show that these neurons preferred similar disparities for ‘‘real’’ and illusory con- were also selective for the orientation of stereoscopic

Figure 6. Stimulus orientation and disparity. (A) Example of an illusory contour stimulus pre- sented at three different orien- tations (from top to bottom: À258,08 +258). The upper rectangle is always presented with crossed (near) disparity, the lower rectangle with zero disparity. The cross indicates an average position of the fixation target of the monkey. (B) Theoretical relationship between stimulus orientation (abscissa) and retinal disparity (ordinate). The horizontal disparity component changes with the cosine (solid line), the vertical component with the sine of the angle of orientation (dotted line). For further explanation, see text.

1022 Journal of Cognitive Neuroscience Volume 14, Number 7

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902320474472 by guest on 03 October 2021 illusory contours. Hence, we recorded orientation response functions to these stimuli. Figure 6A shows an example of an illusory contour stimulus presented at three different orientations. Figure 6B illustrates the theoretical relationship between stimulus orientation and disparity. Since disparity was introduced orthogonal to the ‘‘inducing borders’’ (see above), the stimuli included both horizontal and vertical disparity compo-

nents that varied with stimulus orientation. The hori- Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/7/1018/1757651/089892902320474472.pdf by guest on 18 May 2021 zontal components varied with the cosine (thin solid line) and the vertical components with the sine of the angle of orientation (thin dotted line). Oblique stimuli had therefore horizontal and vertical components that were equal at ±458 (open dots). Horizontal stimuli included only horizontal components (orientation, 08), and vertical stimuli only vertical components (orienta- tion, ±908). Since all neurons that signaled stereoscopic illusory contours also responded to solid contrast borders, we analyzed whether these neurons preferred similar or different orientations for the two types of stimulus. Figure 7A–C shows orientation response functions to illusory contour stimuli (open dots) and to contrast borders (filled dots) for three neurons that signaled stereoscopic illusory contours. The figure shows that both types of stimulus evoked similar response func- tions in these neurons, which implies that they were similarly selective for the orientation of ‘‘real’’ and illusory contours. Also, the response functions were significantly narrower than the cosine of the angle of orientation (A, thin solid line; for details of a statistical comparison, see figure legend). This suggests that these responses reflect selectivity for contour orienta- tion rather than simply a change of the horizontal disparity component.

Human Perception and Cortical Neuron Responses To identify possible correlations between the responses of cortical neurons and human perception, we studied the perception of these contours also in human observ- ers. We used the same types of stimulus as shown in Figures 3 and 6, but without a fixation target. The Figure 7. Orientation selectivity of cortical neurons. Orientation observers were allowed to inspect these stimuli freely, response functions of three neurons to contrast borders (filled dots: but had a frame of reference available in the periphery of A and B, edges; C, light bar) and to illusory contour stimuli (open the visual field (size: 128 Â 128, disparity: zero). They dots). The preferred orientation of these neurons was 08 (A), 128 (B), reported the perceived quality (strength, sharpness) of and 388 (C). The stimuli were presented at 16 different orientations the illusory contour verbally using a given rating scale (Á step, 11.258). The thin solid line in (A) represents a cosine function indicating the relative size of the horizontal disparity (see Methods). In analogy to the neurophysiological component (see Figure 6). In all three neurons, the responses to study in the monkey, we also studied the effects of the illusory contour stimulus were significantly different from a stimulus length, disparity, and orientation. theoretical function following this cosine [A: F(1,15) = 26.9, p < .01; B: F(1,14) = 6.54, p < .05; C: F(1,14) = 17.1, p < .01]. The abscissa scale indicates stimulus orientation (zero indicates horizontal, positive Length values a counterclockwise, negative values a clockwise rotation towards vertical). The ordinate scale indicates the mean normalized Stimulus length was a critical stimulus parameter in per- responses of these neurons recorded during eight cycles of stimulus ception. Figure 8 shows that short stimuli (length: 2–38) movement.

Heider, Spillmann, and Peterhans 1023

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902320474472 by guest on 03 October 2021 induced sharper, more distinct percepts of contour than long stimuli (filled squares, right ordinate scale). The quality of perception weakened with stimulus length, but faint contours were still perceived up to lengths of 6–98. For comparison, this figure also shows the preferred stimulus lengths of the neurons in the monkey visual cortex that signaled stereoscopic illusory contours (histogram, left ordinate scale). The mean

preferred length of these neurons was 3.48 (range: Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/7/1018/1757651/089892902320474472.pdf by guest on 18 May 2021 0.75–8.08, n = 18). Note that neurons recorded in V3/V3A (hatched areas) preferred long stimuli.

Binocular Disparity For a distinct perception of illusory contours, disparity was less critical than stimulus length. Figure 9 illustrates that perception improved with increasing disparity up to 42 min arc and remained constant for larger disparities (filled squares, right ordinate scale). For comparison, Figure 9. Stimulus disparity: Comparison of neuronal responses the figure also includes the preferred stimulus dispari- and human perception. The solid curve indicates the quality of the ties of the neurons in the monkey visual cortex that illusory contour as perceived by human observers. The stimuli were presented with different disparities as indicated on the abscissa signaled these contours (histogram, left ordinate scale). scale. Stimulus orientation was horizontal; the upper rectangle was The mean preferred disparity of these neurons was presented with crossed (near), the lower rectangle with zero disparity. 45 min arc (range: 3–167 min arc, n = 18). Note that Filled squares represent mean responses (±standard errors) of 20 neurons recorded in V3/V3A (hatched area) preferred observers tested with stimuli of seven different lengths (see Methods). large disparities. The histogram shows the preferred stimulus disparity of 18 neurons that signaled stereoscopic illusory contours and for which quantitative measurements were available (open bars: V1, V2; hatched bars: V3/ V3A).

Orientation Figure 10 shows the perceived quality of the illusory contour at different stimulus orientations (filled squares, right ordinate scale). Horizontal orientation (08) and close-to-horizontal orientations induced the strongest percepts of illusory contour, whereas vertical (±908) and close-to-vertical orientations induced only weak percepts. The thin solid line represents a cosine function indicating the relative size of the horizontal dis- parity component at these orientations (see Figure 6B). At first glance, the psychometric function seems to be narrower than this cosine function. However, this im- pression was not supported statistically (see figure legend). This result suggests that horizontal disparity was a critical parameter for the perception of these contours, rather than orientation per se or vertical disparity. For comparison, the figure also includes the Figure 8. Stimulus length: Comparison of neuronal responses and preferred orientations of the neurons in the monkey human perception. The solid curve indicates the quality of the illusory contour as perceived by human observers. The stimuli were presented cortex that signaled stereoscopic illusory contours (his- with different lengths as indicated on the abscissa scale. Stimulus togram, left ordinate scale). These cover the range of orientation was horizontal; the upper rectangle was presented with orientations over which human observers perceived crossed (near), the lower rectangle with zero disparity. Filled squares these contours (mean ±278, range: 6–688, n = 18). represent mean responses (±standard errors) of 20 observers tested at We never found a neuron that preferred vertical or near- 10 different disparities (see Methods). The histogram shows the preferred stimulus lengths of 18 neurons that signaled stereoscopic vertical orientation that could be activated by this stim- illusory contours and for which quantitative measurements were ulus. Neurons that failed to signal stereoscopic illusory available (open bars: V1, V2; hatched bars: V3/V3A). contours showed the full range of possible preferred

1024 Journal of Cognitive Neuroscience Volume 14, Number 7

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902320474472 by guest on 03 October 2021 cross-like object with a uniform surface. The authors found neurons in monkey area V2 that signaled these bars as well as bars defined by luminance contrast. The disparity preferences for the two types of bars were not always related in these neurons. While all neurons preferred crossed (near) disparities for the illusory bar, some preferred crossed (near), others uncrossed (far) disparities for the real bar. In the present paper, the

neurons usually preferred similar disparities for real and Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/7/1018/1757651/089892902320474472.pdf by guest on 18 May 2021 illusory contour stimuli, but exceptions were also found. These results suggest that mechanisms for defining object borders are combined gradually in the visual cortex, involving different types of neurons at different stages of processing. Another finding of the present article concerns the selectivity of V2 neurons for certain figure–ground directions at illusory contours. Neurons that signaled Figure 10. Stimulus orientation: Comparison of neuronal responses stereoscopic illusory contours often preferred one stim- and human perception. The solid curve indicates the quality of the ulus configuration to another, for instance, the upper illusory contour as perceived by human observers. The stimuli were rectangle to be nearer (‘‘figure’’) than the lower rectan- presented at 16 different orientations (Á step, 11.258). The upper gle (‘‘ground’’), and not the reverse. This selectivity rectangle was presented with 42 min arc (crossed) disparity, the lower depended on retinal disparity—the stimulus had to rectangle with zero disparity. Filled squares represent mean responses (±standard errors) of 20 observers tested with stimuli of two different include the neuron’s preferred disparity. We found only lengths (2 and 38). The thin solid line represents a cosine function one neuron that showed this selectivity independent of indicating the relative size of the horizontal disparity component retinal disparity, and this neuron was recorded in V3/ (see Figure 6B). Statistically, the human response function was not V3A. Thus, V2 neurons seem to depend on retinal different from a theoretical function following this cosine [F(1,15) = disparity as has been reported for V1 neurons (Cum- 0.097, p > .05]. The histogram shows the preferred stimulus orientations of 18 neurons that signaled stereoscopic illusory contours ming & Parker, 2000), but in addition, include a mech- and for which quantitative measurements were available (open bars: anism for defining illusory contours and the direction of V1, V2; hatched bars: V3/ V3A). figure and ground at these contours. This agrees with earlier studies on illusory contours not requiring stereo- scopic cues (Baumann, van der Zwan, & Peterhans, orientations, but the overall mean was similar (mean 1997), which showed that V2 neurons can be selective ±298, range: 0–878, n = 68). for certain figure–ground directions at these contours. Recently, comparable properties have been found for V2 neurons that signaled contours from random-dot stereo- DISCUSSION grams (von der Heydt, Zhou, & Friedman, 2000), and in This article describes neuronal responses in the monkey relation to overlapping squares or rectangles defined by visual cortex to a novel stimulus that produces illusory luminance contrast (Zhou, Friedman, & von der Heydt, contours from stereoscopic cues in human perception. 2000). Overall, these results suggest that area V2 The results suggest that these contours are mainly includes several mechanisms for defining contours and represented in the prestriate cortex and that the optimal that some of these also contribute to the segregation of stimulus parameters required by cortical neurons also figure and ground at these contours. produce optimal perception in human observers. Mechanism of Grouping Process Comparison with Other Types of Illusory Contour Spillmann and Werner (1996) noted a common property Stimuli of the different types of illusory contour stimuli—they Earlier studies of neuronal representations of illusory require ‘‘long-range interactions’’ between inducing contours used stimuli that were based on occlusion stimulus elements. Indeed, the perceptual results of cues, but did not require stereopsis (see Introduction). the present paper suggest that faint contours are per- The first study that employed stereoscopic cues com- ceived even for distances of 6–98 between the ‘‘inducing parable to those of the present paper used a stimulus in edges.’’ In previous studies, Peterhans and von der which retinal disparity induced the perception of a Heydt (1991) explained neuronal signals of illusory narrow, vertical bar that appeared to occlude a wider contours by a model that invoked neurons with end- rectangle in the background (Bakin et al., 2000). With- stopped receptive fields (‘‘end-stopped cells’’) to detect out disparity, this stimulus was perceived as a single, occlusion cues (Peterhans & Heitger, 2001; Heitger,

Heider, Spillmann, and Peterhans 1025

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902320474472 by guest on 03 October 2021 von der Heydt, Peterhans, Rosenthaler, & Ku¨bler, 1998). illusory contour stimuli not requiring stereoscopic cues Since the stimuli of the present study are akin to the (Peterhans & von der Heydt, 1989; von der Heydt & illusory bar (or edge) stimuli of these earlier studies, Peterhans, 1989). The higher proportions found by Gro- apart from the fact that they require stereopsis, we call sof et al. (1993) in area V1 were probably related to their upon the same model to explain neuronal signals of stimuli that included elements of luminance contrast at stereoscopic illusory contours. It has been shown that the contour (abutting sine-wave gratings displaced by end-stopped cells can be sensitive to binocular disparity half a cycle), in contrast to the line-gratings used in other as has been reported for neurons with ‘‘end-free’’ studies (von der Heydt et al., 1984; for further discussion,

receptive fields (Peterhans, van der Zwan, & Baumann, see Peterhans, 1997). In the cat visual cortex, area 18 (V2) Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/7/1018/1757651/089892902320474472.pdf by guest on 18 May 2021 1995; Maske, Yamane, & Bishop, 1986). Therefore, we also included higher proportions of neurons sensitive to suggest that these neurons also carry information about illusory contour stimuli than area 17 (V1) (Leventhal & the stereoscopic depth of occlusion cues. In V1, end- Zhou, 1994; Redies et al., 1986). A similar difference stopped cells are mainly located in the superficial between areas V1 and V2 was found for neurons that laminae (2 + 3) that include long-range connections signaled contours from random-dot stereograms; von between neurons preferring similar orientations der Heydt et al. (2000) found 33% (12/36) in V2, and (Gilbert, 1985). These laminae project to area V2 none (0% or 0/26) in V1. These results suggest that area (Felleman & Van Essen, 1991) and could mediate this V2 includes several mechanisms for representing con- information to V2 neurons that signal stereoscopic tours that are not, or rarely implemented in area V1. illusory contours. A wiring scheme within area V2 is also Lamme, Supe`r,and Spekreijse (1998) studied neuro- conceivable—similar proportions of end-stopped cells nal signals of figure–ground segregation in areas V1 and have been found in the two areas (Heider et al., 2000). V2 using stimuli that separated a textured square from a Moreover, V2 neurons usually responded to real and similarly textured ground either by motion cues or illusory contour stimuli, which suggests two inputs to differences in orientation. They also found higher inci- these neurons, one for detecting contrast borders ( bars, dences of these signals in area V2 than in V1, but only in edges), and another for generating signals of illusory awake animals. In anesthetized animals, or animals with contours. Since the stimulus elements of the second V2 lesions, they failed to record V1 signals. These authors input (occlusion cues) are oriented orthogonal to the concluded that V1 signals were not intrinsic but pro- illusory contour and to the contrast borders activating duced by feedback projections from area V2. This fits the first input, this requires connections between neu- with the finding that illusory contour signals have longer rons preferring orthogonal or near-orthogonal orienta- latencies in V1 than in V2 (Lee & Nguyen, 2001). In tions. Such connections have been found in area V2 addition, it has been shown that back projections from (Tamura, Sato, Katsuyama, Hata, & Tsumoto, 1996), but area MT influence figure–ground segregation from seem to be rare in V1 (Gilbert, 1985) and limited to short motion cues in area V1 (Hupe´ et al., 1998). It would be distances (Das & Gilbert, 1999). interesting to study whether analogous feedback projec- In this report, we focused on illusory contours tions also govern the signals of stereoscopic illusory induced by stereoscopic cues and report of neurons contours that we found in area V1. that signal these contours and indicate the figure– ground direction that human observers perceive at Neuronal Responses and Human Perception these contours. Mechanisms involved in the reconstruc- tion of the surfaces associated with these contours, and Apart from the neurophysiological results discussed more global aspects of figure–ground segregation con- above, we also present in this paper psychophysical cerning two-dimensional forms, have not been ad- results, which suggest that the mean stimulus param- dressed. These functions most likely involve more eters required by cortical neurons also induce optimal complex wiring schemes that also include other areas perception of stereoscopic illusory contours in human such as area V4 and the inferotemporal cortex (see, observers. We studied the effects of three stimulus e.g., Grossberg, 1997). parameters on human perception, namely, the effects of stimulus length, disparity, and orientation. With regard to length and disparity, the mean stimulus Anatomical Location parameters preferred by cortical neurons corresponded The anatomical differences between areas V1 and V2, as to those that induced optimal percepts of illusory con- described above, may explain the rare observations of tours in humans. Similarly, the maximum stimulus illusory contour signals in area V1. Bakin et al. (2000) length that still evoked a response in cortical neurons found 8% (1/13) of the V1 neurons to be sensitive to (88) corresponded to the maximal stimulus lengths for illusory bar stimuli, in contrast to the 31% (11/35) found which human observers reported to see illusory con- in area V2. Similarly for the neurons described in the tours. The few neurons recorded in V3/ V3A preferred present paper, only 7% (2/29) were found in area V1, but longer stimuli and larger disparities than the majority of 32% (19/59) in area V2. This difference was similar for the V2 neurons. This finding agrees with the observation

1026 Journal of Cognitive Neuroscience Volume 14, Number 7

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902320474472 by guest on 03 October 2021 that V3/V3A neurons have larger receptive fields than V2 faced, high-resolution oscilloscope screen (Ferranti A5, neurons and often prefer large, crossed or uncrossed peak at 555 nm). These stimuli were projected to the disparities (Adams & Zeki, 2001; Poggio, Gonzales, & eyes of the animal by means of a stereoscope that Krause, 1988; Zeki, 1978). However, since the average produced a fixation distance of 40 cm. Uniform illumi- stimulus parameters required by cortical neurons also nation was added to this display by means of a produced optimal perception in humans, this suggests half-silvered mirror, which produced a background that V2 neurons play an important role in the generation luminance of 22 (or 36) cd/m2 and a stimulus luminance of stereoscopic illusory contours. The results on stim- of 51 (or 72) cd/m2. We recorded the activity of single

ulus orientation suggest that the perception of these neurons extracellularly during the periods of active Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/7/1018/1757651/089892902320474472.pdf by guest on 18 May 2021 contours depends mainly on horizontal disparity and visual fixation using glass-coated platinum–iridium that different orientations are represented by pools of microelectrodes prepared according to Wolbarsht, neurons sensitive to different, comparatively narrow MacNichol, and Wagner (1960), but without platinum- ranges of orientation. black coating. The signals were amplified, fed to ear- phones for listening to the responses, and to a Schmitt trigger for quantitative records and off-line analysis. All METHODS stimuli were also projected onto a separate scope for Physiology plotting receptive fields. For each neuron under study, we determined the optimal stimulus type (bar, edge), its Animal Preparation preferred width, length, and orientation as well as the Three rhesus monkeys (female, body weight 4.5–6.2 kg) size and the position of the receptive field. We deter- were trained on a visual fixation task that reinforced mined the neuron’s sensitivity to binocular disparity foveal viewing. The animals could initiate a trial by using the optimal stimulus and defined its preferred pulling a lever. After an unpredictable time interval disparity and range of sensitivity. The illusory contour (0.5–5 sec), the target (two vertical lines of 1 Â 7 min stimuli were presented as shown in Figures 3 and 6, one arc, separated by 5 min arc) was turned by 908 upon rectangle with the neuron’s preferred disparity, the which the animal had to release the lever within 0.4 sec. other with zero disparity. By listening to the responses, Each correct trial was rewarded with a small drop of fruit we compared the neuronal activity in this condition with juice or water. Experiments began when the animals the activity evoked by monocular stimulation (left and worked with a performance rate of 90–95%. Accuracy of right eye separately) and by a control stimulus in which visual fixation was controlled by means of a TV camera both rectangles were presented with the same disparity. and from the dot displays of the responses. The latter This initial, qualitative testing required about 20–30 min. showed that the scatter of visual fixation was similar Thereafter, we started the quantitative study by record- to that measured by Motter and Poggio (1984) who ing the neuron’s responses in as many of these con- recorded the eye movements under these conditions. ditions as possible. They found random variations of the position of fixation around the target center with standard deviations of Human Perception 6–8 and 7–13 min arc for horizontal and vertical components, respectively. The animals were prepared Perception of illusory contours was tested in two for recording by implanting in succession a head holder groups, each including 20 observers. Group 1 (7 women, and two recording chambers onto the skull, one over 13 men; mean age 33 years; 4 experienced, 16 naı¨ve each hemisphere. These operations were performed observers) evaluated the effect of length and disparity under general anesthesia initiated by a combination of on horizontal contours. Group 2 (4 women, 16 men; ketamine hydrochloride (Ketalar, 5–10 mg/kg, IM) and mean age 31 years; 8 experienced, 12 naı¨ve observers) diazepam (Valium, 0.05–0.1 mg/kg, IM), followed by evaluated contours of two different lengths at different atropine sulfate (0.05–0.1 mg/kg, SC) and pentobarbital orientations. Both groups consisted of undergraduate sodium (Nembutal, 25–30 mg/kg, IP). Anesthesia was students, researchers, and technicians of the University maintained using N2O:O2 (2:1) via a tracheal tube and pen- of Freiburg. All observers had normal or corrected-to- tobarbital sodium as necessary (Nembutal, 2–10 mg/kg, normal vision. The stimuli were generated electronically IV or IP every 1–2 hr). Body temperature, pulse rate, and using Deneba Canvas software and printed on white blood oxygenation were monitored continuously. All paper using an HP LaserJet printer. Stimuli for the left experimentalprocedureswereapprovedbytheVeterinary and right eyes were presented separately to the observ- Office of the Kanton Zurich. er’s eyes by means of a stereoscope (Zeiss, No. 125342) that was fitted with half silvered mirrors and lenses producing a fixation distance of 28.5 cm. Head position Visual Stimulation and Recording was kept constant using an in-built notch for positioning Fixation target and visual stimuli were generated with a the nose. All stimuli were presented at the center of refresh rate of 100 Hz for each eye separately on a flat- the visual field, either light (166 cd/m2)onadark

Heider, Spillmann, and Peterhans 1027

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902320474472 by guest on 03 October 2021 background (7 cd/m2), or reverse. The observers were Felleman, D. J., & Van Essen, D. C. (1991). Distributed instructed to verbally report the subjective quality hierarchical processing in the primate cerebral cortex, Cerebral Cortex, 1, 1–47. (strength, sharpness) of the illusory contour using a ffytche, D. H., & Zeki, S. (1996). Brain activity related to the rating scale from 1 to 10. All stimuli were presented in perception of illusory contours. Neuroimage, 3, 104–108. a predefined, random order. In the first series of experi- Gilbert, C. D. (1985). Horizontal integration in the neocortex. ments (observer group 1), we varied contour length Trends in Neurosciences, 8, 160–165. (2, 3, 4, 5, 6, 7.5, and 98) or disparity (6, 12, 18, 24, 30, 36, Grosof, D. H., Shapley, R. M., & Hawken, M. J. (1993). Macaque V1 neurons can signal ‘‘illusory’’ contours. Nature, 365, 42, 48, 54, and 60 min arc) and kept orientation constant 550–552.

(horizontal). In the second series (observer group 2), we Grossberg, S. (1997). Cortical dynamics of three-dimensional Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/7/1018/1757651/089892902320474472.pdf by guest on 18 May 2021 varied contour orientation and kept the other stimulus figure–ground perception of two-dimensional pictures. parameters constant at optimum values as determined Psychological Review, 104, 618–658. in the first series (length, 2 and 38; disparity, 42 min arc Harris, J. P., & Gregory, R. L. (1973). Fusion and rivalry of illusory contours. Perception, 2, 235–247. introduced orthogonal to the lateral borders of one of Heider, B., Meskenaite, V., & Peterhans, E. (2000). Anatomy the rectangles). In addition, we studied possible effects and physiology of a neural mechanism defining depth order of contrast polarity (light, dark stimuli were alternated), and contrast polarity at illusory contours. European Journal gender, and age on the observers’ responses. Since we of Neuroscience, 12, 4117–4130. failed to find any significant effects, we present in this Heitger, F., von der Heydt, R., Peterhans, E., Rosenthaler, L., &Ku¨bler, O. (1998). Simulation of neural contour paper the pooled results of dark and light stimuli, of mechanisms: Representing anomalous contours. Image and both sexes and all ages. Vision Computing, 16, 407–412. Hirsch, J., DeLaPaz, R. L., Relkin, N. R., Victor, J., Kim, K., Li, T., Borden, P., Rubin, N., & Shapley, R. (1995). Illusory contours Acknowledgments activate specific regions in human visual cortex: Evidence We thank J. Lentjes for technical assistance and F. Stu¨rzel and from functional magnetic resonance imaging. Proceedings of R. Teichmann for help in testing human observers. This the National Academy of Sciences, U.S.A., 92, 6469–6473. research was supported by SNF-SPP grant #5002-044891. Hupe´, J., James, A., Payne, B., Lomber, S., Girard, P., & Bullier, J. (1998). Cortical feedback improves discrimination Reprint requests should be sent to Esther Peterhans, between figure and background by V1, V2 and V3 neurons. Wyderrain 7, CH-3012 Bern, Switzerland, or via e-mail: Nature, 394, 784–787. [email protected]. Kanizsa, G. (1979). Organization in vision. Essays on Gestalt perception. New York: Praeger. Lamme, V. A. F., Supe`r,H., & Spekreijse, H. (1998). REFERENCES Feedforward, horizontal, and feedback processing in the visual cortex. Current Opinion in Neurobiology, 8, Adams, D. L., & Zeki, S. (2001). Functional organization 529–535. of macaque V3 for stereoscopic depth. Journal of Larsson, J., Amunts, K., Gulya´s, B., Malikovic, A., Zilles, K., & Neurophysiology, 86, 2195–2203. Roland, P. E. (1999). Neuronal correlates of real and illusory Anderson, B. L., & Julesz, B. (1995). A theoretical analysis contour perception: Functional anatomy with PET. of illusory contour formation in stereopsis. Psychological European Journal of Neuroscience, 11, 4024–4036. Review, 102, 705–743. Lee, T. S., & Nguyen, M. (2001). Dynamics of subjective Anderson, B. L., & Nakayama, K. (1994). Toward a general contour formation in the early visual cortex. Proceedings of theory of stereopsis: Binocular matching, occluding the National Academy of Sciences, U.S.A., 98, 1907–1911. contours, and fusion. Psychological Review, 101, 414–445. Leventhal, A. G., & Zhou, Y. (1994). Cat visual cortical cells Bakin, J., Nakayama, K., & Gilbert, C. D. (2000). Visual are sensitive to the orientation and direction of ‘‘illusory’’ responses in monkey areas V1 and V2 to three-dimensional contours. Society for Neuroscience Abstracts, 20, 1053. surface configurations. Journal of Neuroscience, 20, Maske, R., Yamane, S., & Bishop, P. O. (1986). End-stopped 8188–8198. cells and binocular depth discrimination in the striate cortex Baumann, R., van der Zwan, R., & Peterhans, E. (1997). Figure– of cats. Proceedings of the Royal Society of London, Series B, ground segregation at contours: A neural mechanism in the 229, 257–276. visual cortex of the alert monkey. European Journal of Mendola, J. D., Dale, A. M., Fischl, B., Liu, A. K., & Tootell, Neuroscience, 9, 1290–1303. R. B. H. (1999). The representation of illusory and real Carman, G. J., & Welch, L. (1992). Three-dimensional illusory contours in human cortical visual areas revealed by contours and surfaces. Nature, 360, 585–587. functional magnetic resonance imaging. Journal of Cumming, B. G., & Parker, A. J. (2000). Local disparity not Neuroscience, 19, 8560–8572. perceived depth is signaled by binocular neurons in cortical Motter, B. C., & Poggio, G. F. (1984). Binocular fixation in area V1 of the macaque. Journal of Neuroscience, 20, the rhesus monkey: Spatial and temporal characteristics. 4758–4767. Experimental Brain Research, 54, 304–314. Das, A., & Gilbert, C. D. (1999). Topography of contextual Nakayama, K. (1996). Binocular visual surface perception. modulations mediated by short-range interactions in Proceedings of the National Academy of Sciences, U.S.A., primary visual cortex. Nature, 399, 655–661. 93, 634–639. Ehrenstein, W. H., & Gillam, B. J. (1998). Early demonstrations Nieder, A., & Wagner, H. (1999). Perception and neuronal of subjective contours, amodal completion, and depth coding of subjective contours in the owl. Nature from half-occlusions: ‘Stereoscopic experiments with Neuroscience, 2, 660–663. silhouettes’ by Adolf von Szily (1921). Perception, 27, Peterhans, E. (1997). Functional organization of area V2 in the 1407–1416. awake monkey. In K. S. Rockland, J. H. Kaas, & A. Peters

1028 Journal of Cognitive Neuroscience Volume 14, Number 7

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902320474472 by guest on 03 October 2021 (Eds.), Cerebral Cortex (vol. 12, pp. 335–357). New York: Sheth, B. R., Sharma, J., Rao, S. C., & Sur, M. (1996). Plenum. Orientation maps of subjective contours in visual cortex. Peterhans, E., & Heitger, F. (2001). Simulation of neuronal Science, 274, 2110–2115. responses defining depth order and contrast polarity at Soriano, M., Spillman, L., & Bach, M. (1996). The abutting illusory contours in monkey area V2. Journal of grating illusion. Vision Research, 36, 109–116. Computational Neuroscience, 10, 195–211. Spillmann, L., & Werner, J. S. (1996). Long-range interactions in Peterhans, E., van der Zwan, R., & Baumann, R. (1995). . Trends in Neurosciences, 10, 428–434. Sensitivity to stereoscopic depth of end-stopped cells in Tamura, H., Sato, H., Katsuyama, N., Hata, Y., & Tsumoto, T. monkey visual cortex. Perception, 24, 6. (1996). Less segregated processing of visual information in Peterhans, E., & von der Heydt, R. (1989). Mechanisms of V2 than in V1 of the monkey visual cortex. European

contour perception in monkey visual cortex: II. Contours Journal of Neuroscience, 8, 300–309. Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/7/1018/1757651/089892902320474472.pdf by guest on 18 May 2021 bridging gaps. Journal of Neuroscience, 9, 1749–1763. Varin, D. (1971). Fenomeni di contrasto e diffusione cromatica Peterhans, E., & von der Heydt, R. (1991). Subjective nell’organizzazione spaziale del campo percettivo. Rivista di contours—bridging the gap between psychophysics and Psicologia, 65, 101–128. physiology. Trends in Neurosciences, 14, 112–119. von der Heydt, R., & Peterhans, E. (1989). Mechanisms of Petry, S., & Meyer, G. E. (1987). The perception of illusory contour perception in monkey visual cortex: I. Lines contours. Berlin: Springer-Verlag. of pattern discontinuity. Journal of Neuroscience, 9, Poggio, G. F., & Fischer, B. (1977). Binocular interaction 1731–1748. and depth sensitivity in striate and prestriate cortex of von der Heydt, R., Peterhans, E., & Baumgartner, G. (1984). behaving rhesus monkey. Journal of Neurophysiology, Illusory contours and cortical neuron responses. Science, 40, 1392–1405. 224, 1260–1262. Poggio, G. F., Gonzales, F., & Krause, F. (1988). Stereoscopic von der Heydt, R., Zhou, H., & Friedman, H. S. (2000). mechanisms in monkey visual cortex: Binocular correlation Representation of stereoscopic edges in monkey visual and disparity selectivity. Journal of Neuroscience, 8, cortex. Vision Research, 40, 1955–1967. 4531–4550. von Szily, A. (1921). Stereoskopische Versuche mit Ramachandran, V. S., & Cavanagh, P. (1985). Subjective Schattenrissen. Albrecht von Graefes Archiv fu¨r contours capture stereopsis. Nature, 317, 527–530. Ophthalmologie, 105, 964–972. Redies, C., Crook, J. M., & Creutzfeldt, O. D. (1986). Neuronal Wolbarsht, M. L., MacNichol, E. F., Jr., & Wagner, H. G. (1960). responses to borders with and without luminance gradients Glass insulated platinum microelectrode. Science, 132, in cat visual cortex and dorsal lateral geniculate nucleus. 1309–1310. Experimental Brain Research, 61, 469–481. Zeki, S. M. (1978). The third visual complex of rhesus monkey Schumann, F. (1900). Beitra¨ge zur Analyse der Gesichtswahr- prestriate cortex. Journal of Physiology, 277, 245–272. nehmungen. Erste Abhandlung. Einige Beobachtungen Zhou, H., Friedman, H. S., & von der Heydt, R. (2000). Coding u¨ber die Zusammenfassung von Gesichtseindru¨cken zu of border ownership in monkey visual cortex. Journal of Einheiten. Zeitschrift fu¨r Psychologie, 23, 1–32. Neuroscience, 20, 6594–6611.

Heider, Spillmann, and Peterhans 1029

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902320474472 by guest on 03 October 2021