Classifying Illusory Contours: Edges Defined by Pacman and Monocular Tokens

Transcription

1 Classifying Illusory Contours: Edges Defined by Pacman and Monocular Tokens GERALD WESTHEIMER AND WU LI Division of Neurobiology, University of California, Berkeley, California Westheimer, Gerald and Wu Li. Classifying illusory contours: edges defined by pacman and monocular tokens. J. Neurophysiol. 77: , Thresholds for the discrimination of orientation were measured in the human fovea for figures and borders delineated by solid lines and by pacman tokens as introduced by Kanizsa, as well as by contours induced by monocular tokens giving a stereoscopic depth illusion of a knife edge. Orientation discrimination of these illusory contours is poorer by a factor of Ç2 than that of equivalent contours made of solid lines and is not much better than that for their supporting structures if taken alone. It is concluded that these kinds of illusory borders do not address the border or edge mechanism in the same way as real lines. Orientation discrimination and simultaneous orientation contrast ( tilt illusion) were compared for a variety of illusory borders. The more robust the borders, i.e., the more sensitive to changes in orientation, the less their susceptibility to the tilt illusion. INTRODUCTION tors ( HP1345A). Pacman tokens were drawn by closely spaced lines of the requisite length. All edges were given orientation changes by recalculating the coordinates of all generating lines according to a rotation around the geometric center of the line or border, or, in the case of the data in Fig. 1, the whole figure. Unfortunately it is impossible to avoid secondary cues in these experiments. If the pacman tokens had remained fixed, rotation of the illusory borders would have necessitated recognizable changes in their internal configuration. Instead we chose to leave the pacman tokens invariant in shape while giving them the combination of rotation and displacement that results from use of the middle of the contour, or the whole figure, as the center of rotation. Whether displacements act as a secondary cue is a question that can be answered independently ( see the DISCUSSION section). The two authors as well as one naive subject served as observers. They faced the monitor screens in a dim room at a distance of 3.75 m, with binocular vision, wearing the needed refractive correction. The experiments whose results are illustrated in Fig. 5 were conducted in a stereoscopic setup, in which each eye saw only its own monitor by means of a system of first-surface mirrors. Thresholds were obtained by the psychophysical method of con- stant stimuli and probit analysis and are expressed in angle of orientation. They indicate the deviation from the vertical for which the observer can correctly identify the direction ( clockwise or counterclockwise) on 75% of occasions. Details are contained in earlier papers from this laboratory ( e.g., Westheimer and Li 1996). Data were gathered in blocks of 140 responses to the same kind of pattern. All threshold measures given in this paper are based on at least two such blocks obtained on different days. No error feed- back was given. Unless otherwise indicated, exposure duration was always 300 ms. A measure of the magnitude of simultaneous orientation contrast ( tilt illusion) was obtained by finding, in the presence of surrounding tilted contours or lines, the actual tilt of the test border when the observer calls it vertical. To obviate bias, it was identified by halving the difference between the 50% point on the probit curves for clockwise and counterclockwise inducing contours. To be perceived as contours, borders need not be drawn out explicitly but can be sketched in by a variety of means. Virtual, subjective, or illusory borders differ in their con- spicuity, but this is not a property easily expressed empiri- cally. On the other hand, orientation, a prominent attribute of a straight border, can be approached via a discrimination threshold, which gives excellent quantitative measures. For example, a simple line just covering the human fovea (Ç30 arcmin) can have the direction of its tilt with respect to the vertical detected when it is half a degree of orientation or less. We have recently used this orientation discrimination threshold to show that some illusory borders are inferior to solid straight lines in several respects: not only do they dis- play higher thresholds, but they take longer to process and are not robust to target motion or masking. In the earlier paper (Westheimer and Li 1996) we studied illusory contours made of orthogonal stacks of lines with and without a gap. Here we extend the investigation to two further types of illusory contours, those delineated by pacman tokens in the manner of Kanizsa figures and the star- tling knife-edge depth borders engendered by monocular features described by Nakayama and Shimojo ( 1990). We then go on to compare the orientation discrimination of a variety of kinds of illusory contours and relate it to their susceptibility to accept an orientation shift induced by sur- rounding tilted lines. METHODS Patterns made of thin lines of medium photopic intensity were created under computer control on high-quality flat display moni- RESULTS In Fig. 1 we show the threshold for orientation discrimination of four squares, all having side lengths of 32 arcmin. If the square has its four sides explicitly drawn, thresholds are good, and the presence of pacman tokens at the corner has no influence. A square outlined only by pacman tokens, which is a well-known Kanizsa figure and paradigmatic of illusory figures, needs about twice as large an orientation difference for threshold discrimination. Occlusion of the open sectors, the traditional way of eliminating the percept of an illusory figure, does not materially affect the threshold. Thus the first and major point of this study has been established: although there may be a compelling percept of an /97 $5.00 Copyright 1997 The American Physiological Society 731

2 732 G. WESTHEIMER AND W. LI illusory square, its orientation cannot be discriminated nearly as well as that of an explicitly drawn square. We analyze the situation further in the experiments illustrated in Fig. 2. Here the orientation discrimination around the vertical has been determined for a variety of edges, all 32 arcmin long. Explicitly drawn edges are best, and pacman delineated edges, occluded or not, are not nearly as good. Then we used two short aligned line segments, 8 arcmin long with 16-arcmin inner separation. These are in fact the vertical contours of the pacman sectors and they demonstrated about the same orientation discrimination when shown alone as when they were part of the illusory configuration. Finally, we gave a 3-arcmin horizontal offset to the pacman tokens and also to the simple line segments. On the whole, taken over the three observers, there was a slight further deterioration of performance. Orientation discrimination of lines improves with line length. To test whether this is also the case for illusory edges outlined by pacman tokens, we obtained the orientation discrimination threshold for three line lengths for equivalent real and illusory edges. As can be seen in Fig. 3, only when there is a real line does the performance improve with increasing length. Illusory contours made of abutting offset lines without luminance gradient have an orientation discrimination FIG. 2. Threshold for the orientation discrimination of a variety of real and illusory edges, all 32 arcmin long: (1) a single line; (2) 2 pacman tokens connected by a line; (3) 2 aligned pacman tokens, in the manner of a Kanizsa figure; (4) 2 aligned pacman tokens with occluded openings; (5) 2 aligned 8-arcmin line segments, providing the same vertical contours as in 2 4 but in isolation; (6) 2 pacman tokens as in 3 but offset horizontally by 3 arcmin; (7) same as 5, but offset horizontally by 3 arcmin. Exposure duration: 300 ms. In conditions 6 and 7, the offset remained fixed throughout the experiment, and the observer s task was to judge the direction of the change in orientation that resulted from rotation of the whole configuration around its middle. threshold that is not only high but also rises further with short exposures ( Westheimer and Li 1996). The question arises, therefore, whether this is also the case for Kanizsa edges. Figure 4 shows that this is not the case. The real and illusory edges to which this report is mainly devoted can have their orientation discriminated just as well for short exposures as for long ones. There is yet a different kind of illusory contour pointed to in detail by Nakayama and Shimojo ( 1990). Panum (1858) and Hering (1862) examined the situation in which there is a monocular contour in the vicinity of a binocular one. The most relevant example is the description by Hering of the fusion of a thick vertical line in one eye with a thin one in the other. There is a compelling percept of a sharp edge with a stereoscopic depth discontinuity. Nakayama and Shimojo used the term knife edge, and we might also de- scribe it as a depth cliff. Nakayama and Shimojo showed that this occurs also when there is just a single monocular line in the presence of binocular features and even when the monocular line is just sketched in by three aligned short vertical line segments. ( There is no stereoscopic depth unless FIG. 1. Threshold for the orientation discrimination around the vertical of (1) a square, (2) a square outlined by both solid lines and 4 pacman tokens, (3) a square outlined by 4 pacman tokens, and (4) a square outlined by 4 pacman tokens with occluded corners. The edges were 32 arcmin long in all cases and the exposure duration was 300 ms. All pacman disks had a diameter of 16 arcmin. To produce orientation changes, the patterns were rotated around their center. Orientation changes of explicitly delineated squares could be detected about twice as well as an illusory figure, and occultation of the latter had no further deteriorating effect.

3 CLASSIFYING ILLUSORY CONTOURS 733 boring tilted contours shown either simultaneously or previously as adapting stimuli ( the tilt aftereffect). We examined the magnitude of this simultaneous orientation contrast in a variety of real and illusory borders by the technique described earlier ( Westheimer 1990). Observers reported whether a given contour appeared tilted clockwise or counterclockwise, but the contours were shown flanked by lines with a 20 tilt, randomly clockwise or counterclockwise in long series of presentations. Results were tallied separately for the two directions of inducing tilt. The two psychometric curves are shifted in the clockwise direction for counterclockwise inducing lines and vice versa; half the difference between their 50% points may be taken to be the mean orientation shift induced by a 20 tilt of the flanks. This measure was obtained for the contours shown in the Fig. 6, inset. Because lengths and distances were the same in all cases, it may be assumed that the numbers represent the susceptibility of these contours to tilt induction. There is an approximately inverse relationship to the orientation discrimination threshold, followed by all contours except those delineated by pacman tokens. FIG. 3. Orientation discrimination thresholds as a function of border length for 2 edges. Solid symbols: edge defined by pacman tokens. Open symbols: explicitly drawn line edge plus pacman tokens. Pacman disks were 16 arcmin in diameter. For solid lines, orientation discrimination, expressed in angles of orientation, progressively improves with line length. The illusory edge does not follow this pattern. Exposure duration: 300 ms. DISCUSSION Figures delineated by illusory contour have a very prominent perceptual presence (see Petry and Meyer 1987 for a comprehensive review; also Spillman and Dresp 1995). They can have induced illusory brightness differences ( Day 1987), are said to be detected in parallel rather than serially (Davis and Driver 1994), and may even seem to capture depth ( Ramachandran and Cavanagh 1985). It is therefore tempting to assign their elaboration to an early stage of visual processing. Luminance increment thresholds in positions occupied by illusory borders do not indicate any special changes that cannot be accounted for by the presence of the inducers ( Dresp and Bonnet 1993), but that does not yet there are at least some binocular features; the phenomenon is not experienced when 1 eye is closed.) We find this illusory depth contour very compelling and wondered whether it conferred additional benefits to orientation discrimination beyond that shown by the broken line alone. Consequently we measured the orientation discrimination threshold for the sketched-in line and for a real line under three conditions: when shown monocularly by themselves, when shown monocularly within a field of random binocular lines, and when the random binocular lines were segregated into right and left depth planes in the manner of what Nakayama and Shimojo call da Vinci stereopsis, i.e., a situation geometrically equivalent to that in which monocularity of the test border is caused by its being shielded from one eye because of occlusion by a partial anterior depth plane. To ensure fully developed stereopsis, a 2-s exposure duration was used in these experiments. The data in Fig. 5 illustrate that in no case does the evoked stereoscopic cliff improve orientation discrimination and, if anything, simulation of da Vinci stereopsis (cases 1 and 4) further impairs performance. This supports our previous conclusion that stereopsis does not yet enter at the neural stage processing contour orientation. Finally we address a related question. All oriented conduration for a variety of edge configurations, all 32 arcmin in length. Thresh- FIG. 4. Orientation discrimination thresholds as a function of exposure figurations are subject to simultaneous orientation contrast olds are substantially invariant with exposure duration in the 50- to 1,000- or, as it is commonly called, the tilt illusion; that is, their ms range except for the illusory contours made up of offset abutting lines perceived orientation is affected by superimposed or neigh- without luminance gradient.

4 734 G. WESTHEIMER AND W. LI outer terminators of the contours that accompanies rotation. Thus, when a vertical line is rotated counterclockwise, is the observer s judgment made on the basis of the line s changed orientation or on the fact that the top end of the line is now displaced to the left with respect to the bottom? Fortunately it is possible to differentiate between these two judgments not merely by instruction to the observers but also by their difference in thresholds. When a line is longer than Ç10 arcmin in the fovea, orientation thresholds for lines are considerably better than position difference thresholds for their two endpoints, so that in such cases the end displacement cue has been eliminated; this also applies to most of the data in this paper. The data in Fig. 3 above constitute an example of this phenomenon that has recently been subjected to a fuller analysis (Westheimer 1996) and that is at the heart of the arguments developed in the present study: because pacman-induced illusory contours have worse orientation thresholds than explicitly drawn lines, they do not utilize fully the orientation-defining neural mechanism. Taken as a whole, the evidence presented in this paper leads one to the conclusion that insofar as judgment of their orientation is concerned, illusory contours induced by Kani- FIG. 5. Orientation discrimination of monocular borders made up of 3 zsa and monocular tokens depend on the fragments of real short lines or a long solid line under 3 conditions. In 1 and 4 the monocular borders that they contain and that perceptual completion or borders are embedded in a field of random binocular short lines with a difference in disparity between the right and left sides; in 2 and 5 all filling in does not manifest itself in the improved perfor- binocularly seen short lines are in the fixation plane, and in 3 and 6 the mance that would be displayed by an equivalent explicit line field is empty except for the monocular border. In all cases there appears stimulus. Moreover, occlusion of the pacman sector ( Figs. a depth edge to 1 side of the monocular features and this edge is particularly 1 and 2, case 4, and Fig. 4, case 2), which abolishes the pronounced when there are many binocular features. The illusory depth sense of an illusory figure, has essentially no additional efborder does not, however, manifest itself in an improved orientation dis- crimination threshold. Exposure duration: 2 s. deny early cortical processing, because it is generally believed that luminance increment thresholds reflect retinal sensitivity and there is no reason to believe that the retina is involved in any direct way in the generation of subjective contours. On the other hand, the orientation attribute of borders is widely believed to have its neural substrate in the primary visual cortex ( Gilbert and Wiesel 1990; Vogels and Orban 1990). Orientation discrimination is a good tool for probing this mechanism, with the general expectation that the more a stimulus addresses the concerned neural apparatus, the smaller the difference in orientation that can be detected. Insofar as different kinds of contours or edges show differences in their orientation discrimination, it may be inferred that they do not have equal access to the apparatus. On the basis of this premise, we conclude that illusory borders of the Kanizsa and monocular types, although undoubtedly prominent perceptually, do not have the benefit of having their orientation processed by the orientation-selective apparatus early in the cortical visual stream. The same applies to the illusory borders generated by offset abutting lines without luminance gradients, which have previously been assigned to a V2 apparatus (Peterhans and von der Heydt 1989; van der Zwan and Wenderoth 1995; von der Heydt and Peterhans 1989). Further evidence for this view is seen in Fig. 3 above, where line length dependency is quite different for real lines and illusory contours. A distinction needs to be drawn between the detection of contour orientation and of the inevitable displacement of the fect. But it would be too facile to make the distinction just between explicit and illusory borders, for there seem to be FIG. 6. Magnitude of the tilt illusion for a variety of real and illusory borders. The length, position, and tilt of the inducing contours were the same in all cases. There is an approximately inverse relationship between the sensitivity to orientation difference and susceptibility to the tilt illusion.

5 CLASSIFYING ILLUSORY CONTOURS 735 real differences between the latter, even when light intensity, and tilt aftereffect has been well established ( Berkley et al. line length, and exposure duration are factored out. In an 1994; Paradiso et al. 1989; Vogels and Orban 1987). But earlier study (Westheimer and Li 1996) we argued that sim- there are indications that they do so to different extents ple illusory contours generated by interrupted lines or (Westheimer 1990). The data in the latter paper and in Fig. stacked terminators of orthogonal lines behave more or less 6 above, for which at least one observer is in common, the same as real lines, if the right conditions of length and permit an ordering of the extent to which tilted inducers can separation are obeyed. The same can probably be said for influence the perceived orientation of the various kinds of the detection of the direction of motion of a single dot (Westheimer illusory contours. The experimental conditions of contour and Wehrhahn 1994). But there is no question that length, position, and orientation of inducing lines are essen- edges delineated by abutting lines with offsets ( Fig. 4, case tially the same in all these studies and allow one to place 5) do not fall into the same category, because not only do illusory contours in a sequence according to decreasing susceptibility they have a higher threshold, but they also take longer to to induced orientation shifts. When this order is process, are more subject to masking, and are not robust compared with the threshold for orientation discrimination, to motion. The illusory contours induced by Kanizsa and it is found that there is a remarkable agreement: the poorer monocular depth tokens studied in this paper fall into an a contour is in orientation discrimination, the more it allows intermediary category, because although their orientation its perceived orientation to be affected by interaction from threshold is higher and does not improve with contour length, surrounding lines ( Fig. 7). And, curiously, the Kanizsa border it is robust to shortening of exposure time and (as we have does not quite fit this order: it shows a smaller tilt illusion ascertained in an experiment not further described here) to than might be expected if the sequences matched exactly. image motion. Moreover, there is a small indication that Perhaps this can be read as another indication that there is some contour completion occurs. Specifically, bringing the indeed a small amount of border completion ( see also Fig. two pacman tokens that define a contour out of alignment 2), but on the whole we have shown that only explicitly by only 3 arcmin (Fig. 2, cases 6 and 7) does raise the drawn lines and closely spaced interrupted contours support threshold in all of our three observers, in a manner that is good orientation discrimination; illusory contours in general reminiscent of the reduction in conspicuity of a Kanizsa share this capability only to an extent that is limited and that triangle when the tokens are misaligned ( Fahle and Koch varies with the type. 1995). Another way of approaching the task of ordering the mag- This research was supported by National Eye Institute Grant EY nitude of the contour attribute of our borders is to exam- Address for reprint requests: G. Westheimer, 321 Life Sciences Addition, ine the effect on them of flanking stimuli of contrasting University of California, Berkeley, CA orientation. That illusory borders can support the tilt illusion Received 28 May 1996; accepted in final form 3 October REFERENCES the sensitivity for orientation discrimination and the magnitude of the tilt illusion: the firmer the border, i.e., the more sensitive it is to perturbations of its orientation, the less its orientation can be influenced. FIG. 7. Rank order for various kinds of illusory contours according BERKLEY, M. A., DEBRUYN, B., AND ORBAN, G. Illusory, motion and luminance-defined contours interact in the human visual system. Vision Res. 34: , DAVIS, G. AND DRIVER, J. Parallel detection of Kanizsa subjective figures in the human visual system. Nature Lond. 371: , DAY, R. H. Cues for edge and the origin of illusory contours: an alternative approach. In: The Perception of Illusory Contours, edited by S. Petry and G. E. Meyer. New York: Springer-Verlag, 1987, p DRESP, B. AND BONNET, C. Psychophysical measures of illusory form perception: further evidence for local mechanism. Vision Res. 33: , FAHLE, M. AND KOCH, C. Spatial displacement, but not temporal asynchrony, destroys figural binding. Vision Res. 35: , GILBERT, C. D. AND WIESEL, T. N. The influence of contextual stimuli on the orientation selectivity of cells in primary visual cortex of cat. Vision Res. 30: , HERING, E. Beitraege zur Physiologie. Leipzig, Germany: Engleman, 1862, p. 25, Fig. 31. NAKAYAMA, K. AND SHIMOJO, S. da Vinci stereopsis: depth and subjective occluding contours from unpaired image points. Vision Res. 30: , PANUM, P.L.Physiologische Untersuchungen ueber das Sehen mit zwei Augen. Kiel, Germany: Schwers, to 2 criteria, sensitivity for orientation discrimination and magnitude of PARADISO, M. A., SHIMOJO, S., AND NAKAYAMA, K. Subjective contours, simultaneous orientation contrast (tilt illusion), on the basis of data from tilt aftereffects, and visual cortical organization. Vision Res. 29: 1205 this paper and from Westheimer (1990), Westheimer and Wehrhahn 1214, (1994), and Westheimer and Li (1996). The experimental conditions of PETERHANS, E. AND VON DER HEYDT, R. Mechanisms of contour perception contour length, foveal vision, exposure duration, position of inducing con- in monkey visual cortex. II. Contours bridging gaps. J. Neurosci. 9: tours, etc., are essentially identical in all experiments and allow the conclusion , to be drawn that there is an approximately inverse relationship between PETRY, S. AND MEYER, G. E. (Editors). The Perception of Illusory Contours. New York: Springer-Verlag, RAMACHANDRAN, V. S. AND CAVANAGH, P. Subjective contours capture stereopsis. Nature Lond. 317: , 1985.

6 736 G. WESTHEIMER AND W. LI SPILLMANN, L. AND DRESP, B. Phenomena of illusory form: can we bridge the gap between levels of explanation? Perception 24: , VAN DER ZWAN, R. AND WENDEROTH, P. Mechanism for purely subjective contour tilt aftereffects. Vision Res. 35: , VOGELS, R. AND ORBAN, G. A. Illusory contour orientation discrimination. Vision Res. 27: , VOGELS, R. AND ORBAN, G. A. How well do response changes in striate neurons signal differences in orientation: a study in the discriminating monkey. J. Neurosci. 10: , VON DER HEYDT, R. AND PETERHANS, E. Mechanism of contour perception in monkey visual cortex. I. Lines of pattern discontinuity. J. Neurosci. 9: , WESTHEIMER, G. Simultaneous orientation contrast for lines in the human fovea. Vision Res. 30: , WESTHEIMER, G. Location and line orientation as distinguishable primitives in spatial vision. Proc. R. Soc. Lond. B Biol. Sci. 263: , WESTHEIMER, G. AND LI, W. Classifying illusory contours by means of orientation discrimination. J. Neurophysiol. 75: , WESTHEIMER, G.AND WEHRHAHN, C. Discrimination of direction of motion in human vision. J. Neurophysiol. 71: 33 37, 1994.