- Email: [email protected]
Subjective contours - bridging the gap between psychophysics and physiology
Subjectivecontours- bridging the gap between psychophysicsandphysiology Esther Peterhansand ROdigervonder Heydt Esther Peterhans and Much is known about the initial stages of visual Rddiger yon der processing up to the striate cortex, but how is visual Heydt are at the information represented and handled at subsequent Dept of Neurology, stages? Phenomena of contour, color and movement University Hospital perception have been used to identify functions of Zi~rich, CH-8091 neurons and to reveal functional differences between Z6rich, Switzerland.
case of color. We are used to thinking that color perception depends on the wavelength composition of light, specifically on the degree to which the three cone types are activated. However, in general, the perceived color of a surface is not determined solely by the light that reaches the eyes from that surface, cortical areas that application of classical receptive-field but depends also on the light reflected from other concepts has not suggested. These differences can be parts of the scene; that is, we speak of 'color contrast' related to theoretical stages of visual processing that and 'induction effects'. Cortical signals that relate to provide stability of perception under changing con- perception of color should be subject to similar ditions of stimulation. effects. Thus the functional role of a signal can be identified from its co-variance with perception under Psychophysical methods have long been used to critical conditions. Co-variance also implies constancy under certain investigate physiological processes that were hidden in the brain and, when physiological data eventually conditions. For example, visual systems are expected became available, they often confLrmed the con- to tend to achieve color constancy. An attribute that clusions drawn from psychophysical experiments, or characterizes the reflectance properties of object indicated a correlation between the physiological and surfaces and remains constant in spite of changes in psychophysical phenomena. The measurements of illumination (which of course affects the light that cone absorption spectra, for example, confirmed the reaches the eyes from a surface) needs to be Young-Helmholtz theory of color vision and demon- established. Psychophysics gives an indication of the degree to which human vision achieves such constrated the physiological correlate of trichromatism. More often, however, particularly in recent decades, stancy. On the basis of this knowledge (or one's own physiological studies have led to findings that neither perception), neural signals can be classified according confirmed expectations nor provided correlates of to whether they show this degree of constancy or not. known psychophysical concepts, such as the recep- By comparing the signals recorded in various parts of tive fields of cortical neurons. Physiology became the visual system, it might be possible to pinpoint the emancipated, so to speak, and was used indepen- site of this processing and eventually understand the dently to study vision. However, many authors have underlying mechanism. There are many other consince studied visual cortical processing without refer- cepts in psychophysics analogous to color that could ence to perception. Indeed, it might appear to be be exploited in this manner; the most promising are pointless to try to find corresponding physiological those that can be related to a theory (such as the goal data for the vast variety of psychophysical phenom- of constancy in the example of color). Thus, the ena. Yet physiology needs to be linked to perception. emphasis is on understanding the physiology and The complexity of the visual cortex defies an approach building theories on physiological observations rather that is comparable to taking a clock apart in order to than looking for correlates of perceptual phenomena. understand its function 1. This article discusses examples of physiological studies that illustrate this idea, rather than attempting Psychophysical concepts as tools in physiology to provide a comprehensive review of studies on The approach that is suggested here is the use of visual cortical processing. First, examples in which perceptual phenomena as tools to make sense of the connection to perception is relatively loose, as in neuronal signals 2'3. In spite of the knowledge that has studies using anesthetized animals, will be discussed. accumulated concerning cortical receptive fields, cor- This will proceed to comparisons of neuronal tical signals are often still enigmatic. This is particu- responses in alert, behaving monkeys with examples larly true for recordings from the extrastriate areas, of human perception, and end with studies in which but even responses from the striate cortex can be the link is most stringent, as in attempts to relate puzzling when stimuli slightly more complex than neuronal responses to the perceptual decisions of a edges and bars are used. Researchers who have not monkey. themselves listened to these signals and tried to understand them might not grasp the difficulty of Color, movement, depth guessing the meaning of the signal that the electrode The relative constancy of color perception has been happens to pick up. The difficulties are accentuated if compared with the responses of cells in areas V1 and recordings are made under more natural conditions, V4 of the monkey 4'5. A collage of various colored for example, in an awake animal. papers was illuminated with light of varied compoSince signals in the 5sual cortex are probably sition. It turned out that, in general, responses of related to perceptual categories, such as form, color, cells in V1 change more with alterations of illumination depth, movement and texture, the co-variance of a than do the perceived colors. For example, responses signal with these aspects of perception can help in of a cell sensitive to long-wave (red) light that was interpreting the signal. This is most apparent in the stimulated by a red paper cease when middle-wave
112
~) 1991, ElsevierSciencePublishersLtd, (UK) 0166- 2236/91/$02.00
TINS, Vol. 14, No. 3, 1991
Fig. 1. The problem of defining occluding contours. (A) Photograph. (B)/Hap of the contrast edges extracted from (.4). The fines of abrupt changes in grey value are depicted, as given by a procedure known as 'Sobel operator' in the field of computer vision and by setting a threshold at 5% of the maximum contrast. Since foreground and background can have equal gray values at the border ('anomalous contour'), the objects are delineated incompletely and spurious connections appear. In (C), the true contours, that is, the lines of discontinuity of depth, are shown and truncations of background structure by occlusion are indicated (light blue shading). This representation corresponds approximately to what we perceive when looking at (A).
(green) light is added to the illumination, even though the paper might still appear red. On the other hand, responses of some cells of V4 co-vary with perception, for example, a red-sensitive V4 cell can still fire under greenish illumination as long as the paper under its receptive field appears red. Thus it seems that the colors, as we perceive them, are first elaborated in V4. The underlying mechanism is still unknown. Although the generalization of these findings could be questioned, this is an important experiment that certainly should be repeated in the alert animal. Objects can move in space, but there is no physical movement in images. The system infers movement from changes of intensity. It is not obvious how this is done, given the direction- and orientation-selective cells in the cortex. For example, when a square moves diagonally up and to the right, the cells stimulated by its edges signal either up or to the right, but none signals the true diagonal motion. But it is the diagonal motion that we perceive. It has been shown with moving gratings and plaids (two superimposed gratings of different orientations) that, in anesthetized monkeys, cells in V1 signal only movements of components, while some cells of area MT (VS) are able to integrate these and signal the movement of the pattern as it is perceived 6. Object movement can only be inferred from the retinal image if the eyes are stationary. In general, however, eye movements must be taken into account. In the extreme, when the eyes track a moving object, its image might not move at all on the retina, but the object is nevertheless perceived as moving. Other non-moving objects are then perceived as stationary, although their images move. Thus, a population of neurons whose responses co-vary with movement perception should (1) signal the direction of movement of a stimulus moving across the receptive field while the eyes are stationary, (2) signal the direction of movement of gaze during tracking when the stimulus remains fixed in the receptive field, and (3) signal no movement when retinal stimulus TINS, Vol. 14, No. 3, 1991
motion is due to an eye movement. In experiments with alert monkeys, it has been found that many cells of area V3A (Ref. 7) and the posterior parietal area 7a (Ref. 8) fulfill postulates (1) and (3), or at least show a reduction in response to eye movement-related stimulus motion, while this was much less frequent in V1 and V2 (Ref. 7). Whether the reduction in response was due to suppression of a motion signal or to its cancellation by an eye movement signal (vector subtraction) is not known. Such cells were called 'realmotion cells '7, although responses during tracking have not been shown. Visual tracking neurons have been described in the posterior parietal cortex and in area MST ~-12. However, it was argued that these play a role in the generation of tracking eye movements rather than in the coding of object movement, because MST lesions produce eye movement deficits 13 and because the receptive field structure of these cells seems inappropriate for discriminating object motion from self motion 12. On the other hand, visual tracking neurons have been shown to signal induced movement s . Note that postulates (2) and (3) do not necessarily require an extraretinal signal since, under certain conditions, the eye movement signal could be derived visually, for example, by assuming that the background is stationary. Neuronal mechanisms that might be appropriate for this have been demonstrated 14-17. Thus, we argue that relatively simple considerations of the rules of perception might provide a useful classification of the motion-related cortical signals. This is independent of the elucidation of the underlying mechanisms. Stereoscopic depth depends on disparity and is fairly independent of the patterns conveying the disparity (for example, we can see depth in randomdot stereograms18). In the alert monkey, simple cells in V1 can signal disparity of isolated bars, but only complex cells can also decode random-dot stereograms 19. Depth perception seems to require an important additional step of processing. When an 113
match (within the limits set by noise of image and detector) are problematic. The photograph in Fig. 1A illustrates this. Figure 1B is a map of the contrast edges extracted from Fig. 1A. It represents the contours only partly. Also 'mergers', such as the connections between the contours of the lamp and the window frame on the left, are found and so are 'deserters', parts of contour that are completely detached from the foreground object and join a background structure to constitute new features, A B C D such as the half-moon shapes to the right of the foreground lamp. Fig, 2. 'Subjective contours'. Simple configurations can produce illusory In Fig. 1C the true contours have been drawn, perception of contours (A), (B) and (D). The contours disappear when lines based on our knowledge of the photographed objects. closing the gaps in the inducing elements are added (C). The subjective The lamp in front is delineated completely, and contours are not necessarily extrapolations of given edges or lines (D). They terminations produced by occlusion are marked by may (A), (B), or may not (D) be accompanied by a brightness illusion. (Taken, light blue shading to indicate that these line endings with permission, from Ref. 31.) and corners are not features of an object, and that the intersected edges or lines are likely to continue object is seen at various distances, the size of its behind the occluding object. This is how the picture is retinal image changes approximately in inverse pro- to be interpreted, and how the scene should be portion to viewing distance, but the span of disparity represented in terms of the 'real' contours of the changes in inverse proportion to the square of viewing photographed objects. This is, in fact, how we see it. distance. Thus, the ratio between span of disparity We do not see the spurious features and false and size of image shrinks when objects move away combinations; instead we see the foreground object from the observer. Nevertheless, they do not seem clearly set apart from the background by a complete to flatten. A book appears to be about as thick at a contour. Does the visual system complete and 'clean distance of 80 cm as at 40 cm, while the disparities up' the contours before we see them? are four times smaller. Our visual system seems to Some simple drawings give the impression of lines scale the disparity signals with viewing distance. or borders bridging gaps (Fig. 2). These 'illusory Viewing distance signals have been demonstrated in contours' have intrigued many investigators z7-zg. In posterior parietal cortex 2° and dependence on stimu- such figures, contour perception is 'illusory' or lus disparity has been demonstrated in many cells of 'subjective', but if one looks at Fig. 1A or the correvisual cortex, but, to our knowledge, no-one has sponding real scene the perceived contours are actually truthful. The phenomenon is clearly the looked for 'real-depth ceils' yet. Findings that show that neuronal signals do not same. We use the term 'anomalous contour '3° to correspond to perception in the expected manner are denote the physical condition when a contour, defined also illuminating. In a checkerboard pattern, for by the two-dimensional projection of a threeexample, the orientation-selective cells of V1 do not, dimensional scene, has vanishing contrast. Extending in general, signal the rows and columns of the checks this definition, a figure is called an 'anomalous-contour we perceive, but rather the diagonals and various figure' if, in principle, it could have been generated in oblique orientations that correspond to the two- such a way. Thus, in contrast to illusory (or subjecdimensional Fourier components of the pattern ex. tive) contours, anomalous contours exist indepenWhile many cells of V1 are selective for spatial dently of whether they can be perceived or not. The frequency, they might not signal the fundamental need for this terminology will become clear in the frequency of low-contrast, square-wave gratings, following discussion. though this is the periodicity perceived 22. Contour coding in the visual c o r t e x The vexing c o n t o u r s Are anomalous contours represented in the visual When a three-dimensional scene is projected onto a cortex? Recordings from alert monkeys have two-dimensional retina, information from near and shown 31-a4 that, while cells in area V1 ignore such distant objects is jumbled. If the system is to contours, many cells in area V2 respond to recognize something, it must detect the contours in anomalous-contour figures as if the contours were the image; that is, the borders that separate the given by edges or lines. These cells fire when an objects from their background. It is often assumed anomalous contour passes over their response field, that contours are defined by contrast, so that the providing its orientation is appropriate (Fig. 3). Their system needs only to look for lines of abrupt changes responses co-vary with perception in several ways. of intensity. Since cells in the primary visual cortex of When intersecting lines are added to the illusory-bar cat and monkey signal orientation and position of figure, responses are reduced or abolished (Fig. 3B contrast borders 23'24, it might be assumed that these and C; compare Fig. 2B and C), and when either half cells define contours. The extraction of contrast of the figure is presented alone, the cell fails to borders is also a fundamental operation in computer respond (not shown). Responses elicited by the vision25,~6. abutting gratings (Fig. 3D) increase with the number Contour definition based on contrast, however, is of line-ends, from zero for one line-end to a maximum inherently incomplete and faulty. Even though fore- for ten line-ends on average, just as the perceptual ground and background intensities can differ statisti- strength of illusory contours depends on the number cally most of the time, the few instances in which they of inducing elements (compare Refs 35, 36). Tilting
\
114
/
() o
0 o
TINS, VoL 14, No. 3, 1991
the lines relative to the contour reduces the responses (compare Refs 37, 38) and shifts the peaks of orientation tuning, just as perceived orientation is altered by oblique lines in the Z611ner and Hering illusions 39. Many cells in V1 are also activated by anomalouscontour figures, but these responses are related to the contrast-defined elements. For example, the lineendings of the abutting gratings (Fig. 3D) can excite end-stopped cells of V1 with receptive fields orthogonal to the anomalous contour. Also, contrary to cells in V2 that respond to figures with gaps of several degrees of visual angle, V1 cells are often sensitive to even tiny gaps in bars or edges 4°. This can be compared with the lack of response to the edges of checkerboards mentioned above 21 and the inhibition produced by contrast-reversed segments in bars 41'42 found in cat area 17. Thus, while the V2 mechanisms tend to bridge gaps, V1 responses specifically imply continuity of contrast borders. It can be seen, then, that V2 cells show interesting new properties that one would never have guessed from their bar and edge responses• It seems that V2 is closer to contour perception than VI• It might come as a surprise that a phenomenon that gave rise to early gestalt psychology27 and that was later regarded as cognitive43 should be evident in monkey cortex• But even cats see illusory contours, as shown in behavioral studies 44'45. Thus, 'cultural differences' do not seem to be a factor. It is surprising to find these signals at a level as low as V2 (in the cat they have even been noted in area 17; Ref. 46, but see also Ref. 47), where the cells still appear to be 'driven' by the stimulus, as is the case in V1. This suggests that we are dealing with 'low-level' processes, although 'top-down' influences cannot easily be ruled out, since the recordings are from alert animals. Cognitive
have been interpreted similarly53. Physiology also predicts that adaptation with anomalous contours in one eye should affect test contours in the other eye (interocular transfer), since only cells of V2 that are nearly always binocular are involved, whereas adaptation with ordinary lines should be only partially effective in transference, since it also involves the monocular populations of V1; this has in fact been confirmed52. Within V2, anomalous-contour responses occur in the thick stripes and the interstripes, Unit 3GD5
A
27.7 It"
B
i"
.-"
•
• ':'.,. ..
" '''~-:,::
.
" ::'.':Z~'..
4.6
C
0.7 D
4.0
contours?
Cognitive theory43'4s assumes that perception results from the system trying to find the most likely explanation for given sensory data. In the case of the illusory triangle, it might consider the possibility that three sectored black discs and three angles happen to be arranged in that particular way or, alternatively, that there is a white triangle overlying three complete discs and an outline triangle - the latter is accepted as the more likely explanation. One argument against the involvement of higher centers is the relatively short latency of the anomalous-contour responses 33 (compare Refs 49, 50). Other arguments come from psychophysical experiments. Prolonged inspection of anomalous contours of one orientation results in subsequently viewed contours of near-by orientations appearing tilted, just as in the tilt after-effect with ordinary lines51.52. However, while adaptation with lines makes other lines and anomalous contours appear tilted, adaptation with anomalous contours only tilts these contours and hardly affects ordinary lines 52. This is precisely what is predicted by the physiological data, since anomalous contour-sensitive cells are only a subset of the orientation-selective cells in V2 (Refs 33, 34). The rest of these remain unadapted and thus provide unbiased orientation signals. Asymmetry found in the training effects of orientation judgement TINS, Vol. 14, No. 3, 1991
i ! I
E
0.0 2°
0.5s (2 °)
Fig. 3. The responses of a cell recorded in area V2 of the monkey visual cortex. While the monkey is fixating the cross, the cell's response field (ellipse) is traversed by a dark bar moving back and forth within a stationary light rectangle (A), by an illusory bar (B) or by a contour produced by discontinuity of lines (D). In (C), the moving notches are closed by thin (2 rain arc) lines. No stimulus is present in (E). The responses of the cell are shown on the right, where each dot corresponds to an action potential, and each row to a cycle of stimulus movement. The forward sweeps are plotted in the left halves of the display and the backward sweeps are plotted in the right halves, with a reversed time axis. Average numbers of action potentials per cycle are shown under each panel. The ceil responds to the bar (,4) as well as to the different anomalous contours (B), (19), but with the closing lines the illusory-bar response is virtually abolished (C). For perception, see Fig. 2B and C. Orientation selectivity was found to be the same for ordinary (A) and anomalous contours (B), (D). While this cell responds weakly to anomalous contours, other cells in V2 respond as strongly as to bars and edges or even more strongly. 115
A
® ®
Y
==I>
B
Fig. 4. A possible mechanism for contour perception. (A) It is assumed that contour cells sum two different kinds of signals that both originate from the same patch of retina - one for detection of contrast borders (grey), and another for detection of pattern discontinuity (light blue), The cell shown in (A) has a vertical orientation. The first type of input comes from simple or complex receptive fields of the same orientation (vertical ellipse) and the second comes from a set of end-stopped receptive fields that are distributed along this orientation, but oriented perpendicular to it (horizontal ellipses with circular disks symbolizing inhibitory end-zones; only four are shown). These cells are thought to respond to corners and line-ends, but not to edges and continual lines. Pairs of end-stopped cells with distant receptive fields are connected by non-linear elements (x) which produce a signal only if both cells are activated simultaneously. In (Z) these signals are summed, together with the first input, to produce the contour signal. (B) The mechanism is activated by the anomalous contour of a triangle and it is turned off when the open sectors are closed by circular line segments. Two corners excite a pair of end-stopped cells; the line segments inhibit them.
but not in the thin stripes of the cytochrome-oxidase pattern s4, where color-coded cells predominate sS. This might relate to the observation that illusory contours fade when the inducing shapes are displayed in equiluminant color contrast 47's6. These findings suggest that mechanisms similar to those in monkey exist in the human visual cortex. Orientation-specific adaptation and disappearance of anomalous contours at equiluminance seem to imply low-level mechanisms.
The theoretical level Physiological data p e r se are not a satisfactory explanation of the phenomenon of illusory contours 57, even though one encounters expressions such as 'physiological explanation' or 'physiological theory' in the literature. The existence of a large number of cells signaling anomalous contours (one-third of the cells of V2, which is the second largest visual area 5s) rather raises the question of why the system puts so much effort into analysing these contours - it can hardly be 116
to create an illusion. As sketched above, the detection of occluding contours is a key problem of understanding images. It seems that, as a result of evolutionary adaptation, the circuitry of V2 can detect these contours at an early stage of processing by the fast, parallel action of a large number of relatively simple modules. The principle of this process relies on the fact that various image features occur in statistical correlation with occluding contours 59, such as the abrupt changes of (1) intensity, (2) structure (see Fig. 1A), (3) binocular disparity, (4) motion and (5) the dynamic occlusion/disocclusion of the background, to list only a few possibilities of what relatively simple mechanisms could cope with. A change in disparity (3) occurs because occluding contours, by definition, are lines of discontinuity of depth. Features (4) and (5) result when the observer or the object moves. In stationary images, only (1), (2) and (3) are available; monocularly, only (1) and (2) are available. Each of these indicators taken by itself is insufficient to reveal the contours (Fig. 1B), but when combined they permit reliable detection.
A possible mechanism Figure 4 shows a simple model of a mechanism that combines evidence from contrast and discontinuity-ofbackground patterns (more sophisticated models including computer simulations exist6°'61). Contrast borders stimulate an input from classical simple or complex cells in the center (grey shading), and terminations of lines and edges excite end-stopped cells with receptive fields distributed along the contour and mainly orthogonal to it (blue shading). Pairs of these cells straddling the center field are connected by gating units to ensure that activation on one side only has no effect (so that illusory contours do not spread out from single corners). These signals are added to that of the classical input. This means that a coincidence of terminations can be equivalent to contrast (compare Ref. 62). In figures like those in Fig. 2, this mechanism is wrong, hence the illusion, but when looking at normal three-dimensional scenes, it can be largely correct. It also produces geometrical illusions when the intersected elements are not orthogonal to the contour, but are tilted 39. This is a consequence of the mainly orthogonal orientation of the end-stopped cells. Clearly, the circuits of lowlevel mechanisms must be rather simple because they have to be replicated many times over for the different locations in the visual field. Illusions are the price of simplicity; speed of parallel processing is the gain. Preattentive parallel processing Can parallel processing of anomalous contours be demonstrated in a visual search task? With ordinary lines one can detect, say, a horizontal target line among vertical lines with a reaction time that is independent of the number of lines presented 63. The same is true when bars made up of anomalous contours of the type shown in Fig. 3D are used 64. A horizontal anomalous bar pops out from an array of vertical anomalous bars. However, since the anomalous contour-sensitive cells 33'34 also respond to contrast borders, anomalous contours do not pop out when they have to compete with other contours (exclusive anomalous-contour detectors might not TINS, Vol. 14, No. 3, 1991
exist). Thus, an illusory triangle produced by a triplet of sectored discs in a group of other triplets of sectored discs cannot be detected by parallel search 65. The simplicity of the proposed mechanism (Fig. 4) should not be taken as indicating that contour perception as a whole is a simple process. As pointed out, one has to assume several different mechanisms acting in parallel, and the manner in which their outputs are combined is sophisticated. For example, binocular disparity not only generates contours 1°'3°, but also interacts with the formation of contours from pattern discontinuities:~°'66: pattern discontinuity produces depth perception by itself 67, or by interaction with motion 6~ and dynamic occlusion/disclosure69'7°. Dealing with situations of occlusion implies that there exists some representation of the hidden parts of objects as sketched in Fig. 1C ('amodal completion'aS'69), which can be demonstrated perceptually 7~. Some of these demonstrations, like those of stereoscopic illusory contours 66, were meant to support a cognitive theory. While these have disclosed much of the complexity of contour processing, it is intriguing to investigate how much cognitive reasoning can be translated into the language of low-level mechanisms ~9. It is important to note that the mechanisms present in visual cortex have evolved to a high degree of sophistication, incorporating a great deal of information about the environment, optical imaging and statistical correlations in images. This implicit knowledge 72 makes the action of these mechanisms appear to be cognitive. It could be said that perceptual hypotheses are 'wired in'. However, the resemblance is superficial. Many demonstrations show that the system does not really evaluate the hypotheses or check for consistency at a global level. Perception can be grossly inconsistent with the known geometry of space (Fig. 5). Certainly, form perception also involves processing at higher levels, and some aspects of illusory contours cannot be explained by stimulus-driven mechanisms alone 73, but the question is whether feedback to lower levels must be assumed.
Strengthening the link between physiology and perception The link with perception is indirect in all the examples discussed above, since neuronal signals and perception were compared in different subjects or even species. Perhaps the most exciting recent development is the introduction of simultaneous recording and perceptual testing. This technique was first used to compare neuronal and perceptual spectral-sensitivity thresholds in the monkey 74. By comparing the neuronal response and perceptual decision for each single-stimulus presentation, it has been shown that variations in response of certain neurons in area MT correlate with the momentary fluctuations of movement perception at threshold 75 and under conditions of binocular rivalry 76, and that microstimulation of small pools of cells hi MT can alter perception dramatically in the way predicted from the motion selectivity of these cells 77. These results demonstrate convincingly that the perception of motion direction is related to the activity in this area and might actually depend on the responses of only small groups of cells. It will be most interesting to see TINS, Vol. 14, No. 3, 1991
Fig. 5. It often seems that contour perception involves 'interpretation' of sensory information in terms of a three-dimensional layout of objects. However, this might simply reflect 'implicit knowledge' in the design of lowlevel cortical mechanisms, Global consistency is not checked, and perception can contradict known geometry of space. (Reproduced, with permission, from Ref. 28.)
whether this paradigm can be extended to more complex perceptual or cognitive functions.
Anatomical segregation of function The stripy pattern of cytochrome-oxidase activity seen in V2 has been interpreted as representing three interlaced systems for color, form and depth processing 55. This intriguing hypothesis is based on the distributions of neuronal selectivity in three sets of stripes for stimulus orientation and length, wavelength and binocular disparity, and on the pattern of connections with V1 (Refs 55, 78, 79). It has received further support from demonstrations of selective connections with the prestriate areas V4 and MT, whose specializations are in accordance with such a scheme s°-s3. However, the segregation of the selectivities in V2 is not quite as clear as might be expected from the proposed segregation of function s4'~5. In fact, this makes sense since perceptual qualities like contour, color, motion and depth are not characterized primarily by unique 'cues', but by their lawful dependence on certain stimulus conditions, including, for example, invariance with changes of contour configuration, illumination, eye movements and viewing distance. Selectivity might not reveal function. Selectivity for disparity, for example, can serve depth as well as form perception TM (see Ref. 80 for further discussion of the multiple roles of 'cues'). Cells in V4 are not more selective for wavelength than are cells in V1 or retinal ganglion cells s6, but they show color constancy4'5. Cells in V2 are not more selective for orientation than are cells in V1, but represent contours in a generalized sense 33'34. Thus, the approach of correlating physiology with perception might give a clearer picture of the courses of cortical processing than the distributions of selectivity. Recent findings in alert monkeys are encouraging. Anomalous-contour responses are found in interstripes and thick stripes, but not in thin stripes (thought to process color), while orientation 117
Acknowledgements We thank Dorothea Wemger for correcting the English, Friedrich Heitger and Lukas Rosenthaler for providing the computed contrast borders of Fig. 18, and Ernst Peterhans for modeling the background lamp. We also thank Peter O. Bishop and Peter Hammond for comments and critictsm of an earlier version of the manuscript, Supported by Swiss NF Grant 3 939 84.
selectivity is common even in thin stripes and length selectivity is equally frequent in all three sets 54. Indications of shape-from-motion processing were found to be most frequent in the thick stripes, while direction selectivity was equally rare in all three sets of stripes 4°'87. The question is, what is achieved at the higher stages of cortical processing, rather than how much of the receptive-field properties of the lower stages is preserved?
Concluding remarks Physiological studies that explicitly used perceptual phenomena for interpreting neuronal signals have been discussed. These seem to mark a promising new trend in vision research. A more representative account of the attempts to relate perception to physiology can be found in Ref. 88. In addition, the discussion of the psychophysical and theoretical studies on contour processing is limited, and thus biased by the scope of this review. Contours are necessary, even if they are subjective.
Selected references 1 Marr, D. (1980) Phil. Trans. R. Soc. London Ser. B 290, 199-218 2 Baumgartner, G. (1986) in Diseases of the Nervous System: Clinical Neurobiology (Vol. 2) (Asbury, A. K., McKann, G. M. and McDonald, W. I., eds), pp. 804-815, Saunders 3 yon der Heydt, R. (1987) Rev. Physiol. Biochem. Pharmacol. 108, 69-150 4 Zeki, S. (1983) Neuroscience 9, 741-765 5 Zeki, S. (1983) Neuroscience 9, 767-781 6 Movshon, J. A., Adelson, E. H., Gizzi, M. S. and Newsome, W. T. (1985) Exp. Brain Res. (Suppl. 11), 117-151 7 Galletti, C., Battaglini, P. P. and Fattori, P. Exp. Brain Res. (in press) 8 Sakata, H., Shibutani, H., Kawano, K. and Harrington, T. L. (1985) Vision Res. 25,453-463 9 Sakata, H., Shibutani, H. and Kawano, K. (1983) J. Neurophysiol. 49, 1364-1380 10 Komatsu, H. and Wurtz, R. H. (1988) J. Neurophysiol. 60, 580-603 11 Newsome, W. T., Wurtz, R. H. and Komatsu, H. (1988) J. Neurophysiol. 60, 604-620 12 Komatsu, H. and Wurtz, R. H, (1988) J. Neurophysiol. 60, 621-644 13 DLirsteler, M. R. and Wurtz, R. H. (1988)J. Neurophysiol. 60, 940-965 14 AIIman, J., Miezin, F. and McGuinness, E. (1985) Perception 14, 105-126 15 Tanaka, K. et al. (1986)J. NeuroscL 6, 134-144 16 Frost, B. J., Scilly, P. L. and Wong, S. C. P. (1981) Exp. Brain Res. 43, 173-185 17 Orban, G. A., Gulyas, B. and Vogels, R. (1987) J. Neurophysiol. 57, 1792-1812 18 Julesz, B. (1971) Foundations of Cyclopean Perception, University of Chicago Press 19 Poggio, G. F. (1984) in Dynamic Aspects of Neocortical Function (Edelman, G. M., Gall, W. E. and Cowan, W. M., eds), pp. 613-635, Wiley 20 Sakata, H., Shibutani, H. and Kawano, K. (1980) J. Neurophysiol. 43, 1654-1672 21 De Valois, K. K., De Valois, R. L. and Yund, E. W. (1979) J. Physiol. 291,483-505 22 Maffei, L., Morrone, C., Pirchio, M. and Sandini, G. (1979) J. Physiol. 296, 27-47 23 Hubel, D. H. and Wiesel, T. N. (1962) J. Physiol. 160, 106-154 24 Hubel, D. H. and Wiesel, T. N. (1968) J. Physiol. 195, 21 5-243 25 Marr, D. and Hildreth, E. (1980) Proc. R. Soc. London Ser. B 207, 187-217 26 Poggio, T., Gamble, E. B, and Little, J. J. (1988)Science 242, 436-440 27 Schumann, F. (1900) Z. Psychol. 23, 1-32 28 Kanizsa, G. (1979) Organization in Vision. Essays on Gestalt 118
Perception, Praeger 29 Petry, S. and Meyer, G. E. (1987) The Perception of Illusory Contours, Springer 30 Lawson, R. B. and Gulick, W. L. (1967) Vision Res. 7, 271-297 31 vonder Heydt, R., Peterhans, E. and Baumgartner, G. (1984) Science 224, 1260-1262 32 Peterhans, E., yon der Heydt, R. and Baurngartner, G. (1986) in Visual Neuroscience (Pettigrew, J. D., Sanderson, K. J. and Levick, W. R., eds), pp. 343-351, Cambridge University Press 33 yon der Heydt, R. and Peterhans, E. (1989) J. Neurosci. 9, 1731-1748 34 Peterhans, E. and vonder Heydt, R. (1989) J. Neurosd. 9, 1749--1763 35 Petry, S., Harbeck, A., Conway, J. and Levey, J. (1983) Percept. Psychophys. 34, 169--174 36 Fuld, K. and O'Donnell, K. (1984) in Sensory Experience, Adaptation, and Perception. Festschrift for Ivo Kohler (Spillmann, L. and Wooten, B. R., eds), pp. 461-469, Lawrence Erlbaum Associates 37 Spillmann, L. (1975) in Gestalttheorie der Modernen Psychologie (Ertel, S., Kemmler, L. and Stadler, M., eds), pp. 210-218, Steinkopf 38 Kennedy, J. M. (1978) Perception 7, 605-607 39 yon der Heydt, R. and Peterhans, E. (1989) in Neural Mechanisms of Visual Perception (Lam, D. M. K. and Gilbert, C. D., eds), pp. 157-170, Portfolio Publishing 40 Peterhans, E. and vonder Heydt, R. (1989) in Seeing Contour and Colour (Kulikowski, J. J., Dickinson, C. M. and Murray, I. J., eds), pp. 125-130, Pergamon 41 Hammond, P. and MacKay, D. M. (1983) J. Physiol. 337, 69-87 42 Hammond, P. and MacKay, D. M. (1985) J. Physiol. 359, 315-329 43 Gregory, R. L. (1972) Nature 238, 51-52 44 Bravo, M., Blake, R. and Morrison, 5. (1988) Vision Res. 28, 861-865 45 De Weerd, P., Vandenbussche, E., De Bruyn, B. and Orban, G. A. (1990) Behav. Brain Res. 39, 1-17 46 Redies, C., Crook, J. M. and Creutzfeldt, O. D. (1986) Exp. Brain Res. 61,469-481 47 Gregory, R. L. (1987) in The Perception of Illusory Contours (Petry, S. and Meyer, G. E., eds), pp. 81-89, Springer 48 Rock, I. and Anson, R. (1979) Perception 8, 665-4581 49 Reynolds, R. I. (1981) Perception 10, 107-115 50 Gellatly, A. R. H. (1980) Perception 9, 599-602 51 Smith, A. and Over, R. (1975) Nature 257, 581-582 52 Paradiso, M. A., Shimojo, S. and Nakayama, K. (1989) Vision Res. 29, 1205-1213 53 Vogels, R. and Orban, G. A. (1987) Vision Res. 27, 453-467 54 Peterhans, E. and von clef Heydt, R. (1989) Soc. Neurosci. Abstr. 15, 161 55 Hubel, D. H. and Livingstone, M. S. (1987) J. Neurosci. 7, 3378-3415 56 Brigner, W. L. and Gallagher, M. B. (1974) Perceptual and Motor Skills 38, 1047-1053 57 Marr, D. (1982) Vision. A Computational Investigation into the Human Representation and Processing of Visual Information, Freeman 58 Van Essen, D. C. (1985)in Cerebral Cortex (Vol. 3) (Peters, A. and Jones, E. G., eds), pp. 259-329, Plenum 59 von der Heydt, R. and Peterhans, E. (1989) in Seeing Contour and Colour (Kulikowski, J. J., Dickinson, C. M. and Murray, I. J., eds), pp. 729-734, Pergamon 60 Grossberg, S. and Mingolla, E. (1987) in The Perception of Illusory Contours (Petry, S. and Meyer, G. E., eds), pp. 116-125, Springer 61 Finkel, L. H. and Edelman, G. M. (1989) J. Neurosci. 9, 3188-3208 62 Spillmann, L., Fuld, K. and Neumeyer, C. (1984) Perception 13,513-520 63 Treisman, A. (1988) Q. J. Exp. Psychol. (A) 40, 201-237 64 Gurnsey, R. and Humphrey, G. K. (1990) Invest. Ophthalmol. Visual Sci. 31, 107 (ARVO suppl.) 65 Grabowecky, M. and Treisman, A. (1989) Invest. Ophthalmol. Visual Sci. 30, 457 (ARVO suppl.) 66 Harris, J. P. and Gregory, R. L. (1973) Perception 2,235-247 67 Coren, S. (1972) Psychol. Rev. 79, 359-367 68 Yonas, A., Craton, L. G. and Thompson, W. B. (1987) Percept. Psychophys. 41, 53-59 69 Michotte, A., Thines, G. and Crabbe, G. (1964) Les Complements Amodaux des Structures Perceptives (Studia psychoTINS, Vol. 14, No. 3, 1991
Iogica), Publications Universitaires de Louvain 70 Kaplan, G. A. (1969) PercepL Psychophys. 4, 193-198 71 Nakayama, K. and Shimojo, S. (1989)in NeuralMechanisms of Visual Perception (Lain, D. M. K. and Gilbert, C. D., eds), pp. 281-296, Portfolio Publishing 72 De Weert, C. M. M. and van Kruysbergen, N. A. W. H. (1987) in The Perception of Illusory Contours (Petry, S. and Meyer, G. E., eds), pp. 165-170, Springer 73 Bradley, D. R. and Dumais, S. T. (1975) Nature257, 582-584 74 Sperling, H. G., Crawford, M. L. J. and Espinoza, S. (1978) Mod. ProbL Ophthalmol. 19, 2-18 75 Newsome, W. T., Britten, K. H., Movshon, J. A. and Shadlen, M. (1989) in Neural Mechanisms of Visual Perception (Lam, D. M. K. and Gilbert, C. D., eds), pp. 171-198, Portfolio Publishing 76 Logothetis, N. K. and Schall, J. D. (1989) in Neural Mechanisms of Visual Perception (Lain, D. M. K. and Gilbert, C. D., eds), pp. 199-222, Portfolio Publishing 77 Salzman, C. D., Britten, K. H. and Newsome, W. T. (1990)
Invest. OphthalmoL Vision Sci. 31,238 (ARVO suppl.) 78 Livingstone, M. S. and Hubel, D. H. (1984) J. NeuroscL 4, 309-356 79 Livingstone, M. S. and Hubel, D. M. (1987) J. Neurosci. 7, 3371-3377 80 DeYoe, E. A. and Van Essen, D. C. (1988) Trends Neurosci. 11,219-226 81 Zeki, S. and Shipp, S. (1988) Nature 335, 311-317 82 Shipp, S. and Zeki, S. (1989) Eur. J. Neurosci. 1,333-354 83 Zeki, S. and Shipp, S. (1989) Eur. J. Neurosci. 1,494-506 84 DeYoe, E. A. and Van Essen,D. C. (1985) Nature 317, 58--61 85 Levitt, J. B. and Movshon, J. A. (1990) Invest. Ophthalmol. Vision Sci. 31, 89 (ARVO suppl.) 86 Schein, S. J., Marrocco, R. T. and De Monasterio, F. M. (1982) J. Neurophysiol. 47, 193-213 87 Peterhans, E. and vonder Heydt, R. (1989) InvesL Ophthalmol. Vision ScL 30, 299 88 Spillmann, L. and Werner, J. S. (1990) Visual Perception The Neurophysiological Foundations, Academic
ERRATUM In the article 'Dopaminergic innervation of the cerebral cortex: unexpected differences between rodents and primates' by B. Berger, P. Gaspar and C. Verney (January 1991, Vol. 14, pp. 21-27) an error was introduced to the colour shading of part of Fig. 1. The DA cell groups A9 and A10 shown in the section at the bottom of the Figure were given the wrong shades of blue. We reproduce the correct Figure here and apologise to the authors and readers for this error.
par
v
rs
$m
tern
c i n g ~ ent
Pf / ~ . ~ . ~ i
6
ns 1
RAT 3turn
,.--, "-~
~''
• "
MACAQUE 10ram
A10-A9
European NeuroscienceConference A meeting on K÷ channels and neuronal mechanisms will be held at Chateau de Fillerval, Mouy, France from 24-28 November, 1991. A maximum of 65 attendees, in addition to 25 invited speakers, will be selected. All expenses for participants from Eastern European countries will be met by the organizers. In addition, one or more fellowships will be offered to Eastern European participants on a competitive basis to enable them to visit a Western European laboratory involved in research on the topic of the meeting. Interested applicants should contact the organizer, Dr Y. Ben-Ari, INSERM U-29, HOpital de Port-Royal, 123 Bd de Port-Royal, 75674 Paris Cedex 14, France as soon as possible with a brief c.v., publication list and area of interest (two pages maximum). TINS, Vol. 14, No. 3, 1991
119
case of color. We are used to thinking that color perception depends on the wavelength composition of light, specifically on the degree to which the three cone types are activated. However, in general, the perceived color of a surface is not determined solely by the light that reaches the eyes from that surface, cortical areas that application of classical receptive-field but depends also on the light reflected from other concepts has not suggested. These differences can be parts of the scene; that is, we speak of 'color contrast' related to theoretical stages of visual processing that and 'induction effects'. Cortical signals that relate to provide stability of perception under changing con- perception of color should be subject to similar ditions of stimulation. effects. Thus the functional role of a signal can be identified from its co-variance with perception under Psychophysical methods have long been used to critical conditions. Co-variance also implies constancy under certain investigate physiological processes that were hidden in the brain and, when physiological data eventually conditions. For example, visual systems are expected became available, they often confLrmed the con- to tend to achieve color constancy. An attribute that clusions drawn from psychophysical experiments, or characterizes the reflectance properties of object indicated a correlation between the physiological and surfaces and remains constant in spite of changes in psychophysical phenomena. The measurements of illumination (which of course affects the light that cone absorption spectra, for example, confirmed the reaches the eyes from a surface) needs to be Young-Helmholtz theory of color vision and demon- established. Psychophysics gives an indication of the degree to which human vision achieves such constrated the physiological correlate of trichromatism. More often, however, particularly in recent decades, stancy. On the basis of this knowledge (or one's own physiological studies have led to findings that neither perception), neural signals can be classified according confirmed expectations nor provided correlates of to whether they show this degree of constancy or not. known psychophysical concepts, such as the recep- By comparing the signals recorded in various parts of tive fields of cortical neurons. Physiology became the visual system, it might be possible to pinpoint the emancipated, so to speak, and was used indepen- site of this processing and eventually understand the dently to study vision. However, many authors have underlying mechanism. There are many other consince studied visual cortical processing without refer- cepts in psychophysics analogous to color that could ence to perception. Indeed, it might appear to be be exploited in this manner; the most promising are pointless to try to find corresponding physiological those that can be related to a theory (such as the goal data for the vast variety of psychophysical phenom- of constancy in the example of color). Thus, the ena. Yet physiology needs to be linked to perception. emphasis is on understanding the physiology and The complexity of the visual cortex defies an approach building theories on physiological observations rather that is comparable to taking a clock apart in order to than looking for correlates of perceptual phenomena. understand its function 1. This article discusses examples of physiological studies that illustrate this idea, rather than attempting Psychophysical concepts as tools in physiology to provide a comprehensive review of studies on The approach that is suggested here is the use of visual cortical processing. First, examples in which perceptual phenomena as tools to make sense of the connection to perception is relatively loose, as in neuronal signals 2'3. In spite of the knowledge that has studies using anesthetized animals, will be discussed. accumulated concerning cortical receptive fields, cor- This will proceed to comparisons of neuronal tical signals are often still enigmatic. This is particu- responses in alert, behaving monkeys with examples larly true for recordings from the extrastriate areas, of human perception, and end with studies in which but even responses from the striate cortex can be the link is most stringent, as in attempts to relate puzzling when stimuli slightly more complex than neuronal responses to the perceptual decisions of a edges and bars are used. Researchers who have not monkey. themselves listened to these signals and tried to understand them might not grasp the difficulty of Color, movement, depth guessing the meaning of the signal that the electrode The relative constancy of color perception has been happens to pick up. The difficulties are accentuated if compared with the responses of cells in areas V1 and recordings are made under more natural conditions, V4 of the monkey 4'5. A collage of various colored for example, in an awake animal. papers was illuminated with light of varied compoSince signals in the 5sual cortex are probably sition. It turned out that, in general, responses of related to perceptual categories, such as form, color, cells in V1 change more with alterations of illumination depth, movement and texture, the co-variance of a than do the perceived colors. For example, responses signal with these aspects of perception can help in of a cell sensitive to long-wave (red) light that was interpreting the signal. This is most apparent in the stimulated by a red paper cease when middle-wave
112
~) 1991, ElsevierSciencePublishersLtd, (UK) 0166- 2236/91/$02.00
TINS, Vol. 14, No. 3, 1991
Fig. 1. The problem of defining occluding contours. (A) Photograph. (B)/Hap of the contrast edges extracted from (.4). The fines of abrupt changes in grey value are depicted, as given by a procedure known as 'Sobel operator' in the field of computer vision and by setting a threshold at 5% of the maximum contrast. Since foreground and background can have equal gray values at the border ('anomalous contour'), the objects are delineated incompletely and spurious connections appear. In (C), the true contours, that is, the lines of discontinuity of depth, are shown and truncations of background structure by occlusion are indicated (light blue shading). This representation corresponds approximately to what we perceive when looking at (A).
(green) light is added to the illumination, even though the paper might still appear red. On the other hand, responses of some cells of V4 co-vary with perception, for example, a red-sensitive V4 cell can still fire under greenish illumination as long as the paper under its receptive field appears red. Thus it seems that the colors, as we perceive them, are first elaborated in V4. The underlying mechanism is still unknown. Although the generalization of these findings could be questioned, this is an important experiment that certainly should be repeated in the alert animal. Objects can move in space, but there is no physical movement in images. The system infers movement from changes of intensity. It is not obvious how this is done, given the direction- and orientation-selective cells in the cortex. For example, when a square moves diagonally up and to the right, the cells stimulated by its edges signal either up or to the right, but none signals the true diagonal motion. But it is the diagonal motion that we perceive. It has been shown with moving gratings and plaids (two superimposed gratings of different orientations) that, in anesthetized monkeys, cells in V1 signal only movements of components, while some cells of area MT (VS) are able to integrate these and signal the movement of the pattern as it is perceived 6. Object movement can only be inferred from the retinal image if the eyes are stationary. In general, however, eye movements must be taken into account. In the extreme, when the eyes track a moving object, its image might not move at all on the retina, but the object is nevertheless perceived as moving. Other non-moving objects are then perceived as stationary, although their images move. Thus, a population of neurons whose responses co-vary with movement perception should (1) signal the direction of movement of a stimulus moving across the receptive field while the eyes are stationary, (2) signal the direction of movement of gaze during tracking when the stimulus remains fixed in the receptive field, and (3) signal no movement when retinal stimulus TINS, Vol. 14, No. 3, 1991
motion is due to an eye movement. In experiments with alert monkeys, it has been found that many cells of area V3A (Ref. 7) and the posterior parietal area 7a (Ref. 8) fulfill postulates (1) and (3), or at least show a reduction in response to eye movement-related stimulus motion, while this was much less frequent in V1 and V2 (Ref. 7). Whether the reduction in response was due to suppression of a motion signal or to its cancellation by an eye movement signal (vector subtraction) is not known. Such cells were called 'realmotion cells '7, although responses during tracking have not been shown. Visual tracking neurons have been described in the posterior parietal cortex and in area MST ~-12. However, it was argued that these play a role in the generation of tracking eye movements rather than in the coding of object movement, because MST lesions produce eye movement deficits 13 and because the receptive field structure of these cells seems inappropriate for discriminating object motion from self motion 12. On the other hand, visual tracking neurons have been shown to signal induced movement s . Note that postulates (2) and (3) do not necessarily require an extraretinal signal since, under certain conditions, the eye movement signal could be derived visually, for example, by assuming that the background is stationary. Neuronal mechanisms that might be appropriate for this have been demonstrated 14-17. Thus, we argue that relatively simple considerations of the rules of perception might provide a useful classification of the motion-related cortical signals. This is independent of the elucidation of the underlying mechanisms. Stereoscopic depth depends on disparity and is fairly independent of the patterns conveying the disparity (for example, we can see depth in randomdot stereograms18). In the alert monkey, simple cells in V1 can signal disparity of isolated bars, but only complex cells can also decode random-dot stereograms 19. Depth perception seems to require an important additional step of processing. When an 113
match (within the limits set by noise of image and detector) are problematic. The photograph in Fig. 1A illustrates this. Figure 1B is a map of the contrast edges extracted from Fig. 1A. It represents the contours only partly. Also 'mergers', such as the connections between the contours of the lamp and the window frame on the left, are found and so are 'deserters', parts of contour that are completely detached from the foreground object and join a background structure to constitute new features, A B C D such as the half-moon shapes to the right of the foreground lamp. Fig, 2. 'Subjective contours'. Simple configurations can produce illusory In Fig. 1C the true contours have been drawn, perception of contours (A), (B) and (D). The contours disappear when lines based on our knowledge of the photographed objects. closing the gaps in the inducing elements are added (C). The subjective The lamp in front is delineated completely, and contours are not necessarily extrapolations of given edges or lines (D). They terminations produced by occlusion are marked by may (A), (B), or may not (D) be accompanied by a brightness illusion. (Taken, light blue shading to indicate that these line endings with permission, from Ref. 31.) and corners are not features of an object, and that the intersected edges or lines are likely to continue object is seen at various distances, the size of its behind the occluding object. This is how the picture is retinal image changes approximately in inverse pro- to be interpreted, and how the scene should be portion to viewing distance, but the span of disparity represented in terms of the 'real' contours of the changes in inverse proportion to the square of viewing photographed objects. This is, in fact, how we see it. distance. Thus, the ratio between span of disparity We do not see the spurious features and false and size of image shrinks when objects move away combinations; instead we see the foreground object from the observer. Nevertheless, they do not seem clearly set apart from the background by a complete to flatten. A book appears to be about as thick at a contour. Does the visual system complete and 'clean distance of 80 cm as at 40 cm, while the disparities up' the contours before we see them? are four times smaller. Our visual system seems to Some simple drawings give the impression of lines scale the disparity signals with viewing distance. or borders bridging gaps (Fig. 2). These 'illusory Viewing distance signals have been demonstrated in contours' have intrigued many investigators z7-zg. In posterior parietal cortex 2° and dependence on stimu- such figures, contour perception is 'illusory' or lus disparity has been demonstrated in many cells of 'subjective', but if one looks at Fig. 1A or the correvisual cortex, but, to our knowledge, no-one has sponding real scene the perceived contours are actually truthful. The phenomenon is clearly the looked for 'real-depth ceils' yet. Findings that show that neuronal signals do not same. We use the term 'anomalous contour '3° to correspond to perception in the expected manner are denote the physical condition when a contour, defined also illuminating. In a checkerboard pattern, for by the two-dimensional projection of a threeexample, the orientation-selective cells of V1 do not, dimensional scene, has vanishing contrast. Extending in general, signal the rows and columns of the checks this definition, a figure is called an 'anomalous-contour we perceive, but rather the diagonals and various figure' if, in principle, it could have been generated in oblique orientations that correspond to the two- such a way. Thus, in contrast to illusory (or subjecdimensional Fourier components of the pattern ex. tive) contours, anomalous contours exist indepenWhile many cells of V1 are selective for spatial dently of whether they can be perceived or not. The frequency, they might not signal the fundamental need for this terminology will become clear in the frequency of low-contrast, square-wave gratings, following discussion. though this is the periodicity perceived 22. Contour coding in the visual c o r t e x The vexing c o n t o u r s Are anomalous contours represented in the visual When a three-dimensional scene is projected onto a cortex? Recordings from alert monkeys have two-dimensional retina, information from near and shown 31-a4 that, while cells in area V1 ignore such distant objects is jumbled. If the system is to contours, many cells in area V2 respond to recognize something, it must detect the contours in anomalous-contour figures as if the contours were the image; that is, the borders that separate the given by edges or lines. These cells fire when an objects from their background. It is often assumed anomalous contour passes over their response field, that contours are defined by contrast, so that the providing its orientation is appropriate (Fig. 3). Their system needs only to look for lines of abrupt changes responses co-vary with perception in several ways. of intensity. Since cells in the primary visual cortex of When intersecting lines are added to the illusory-bar cat and monkey signal orientation and position of figure, responses are reduced or abolished (Fig. 3B contrast borders 23'24, it might be assumed that these and C; compare Fig. 2B and C), and when either half cells define contours. The extraction of contrast of the figure is presented alone, the cell fails to borders is also a fundamental operation in computer respond (not shown). Responses elicited by the vision25,~6. abutting gratings (Fig. 3D) increase with the number Contour definition based on contrast, however, is of line-ends, from zero for one line-end to a maximum inherently incomplete and faulty. Even though fore- for ten line-ends on average, just as the perceptual ground and background intensities can differ statisti- strength of illusory contours depends on the number cally most of the time, the few instances in which they of inducing elements (compare Refs 35, 36). Tilting
\
114
/
() o
0 o
TINS, VoL 14, No. 3, 1991
the lines relative to the contour reduces the responses (compare Refs 37, 38) and shifts the peaks of orientation tuning, just as perceived orientation is altered by oblique lines in the Z611ner and Hering illusions 39. Many cells in V1 are also activated by anomalouscontour figures, but these responses are related to the contrast-defined elements. For example, the lineendings of the abutting gratings (Fig. 3D) can excite end-stopped cells of V1 with receptive fields orthogonal to the anomalous contour. Also, contrary to cells in V2 that respond to figures with gaps of several degrees of visual angle, V1 cells are often sensitive to even tiny gaps in bars or edges 4°. This can be compared with the lack of response to the edges of checkerboards mentioned above 21 and the inhibition produced by contrast-reversed segments in bars 41'42 found in cat area 17. Thus, while the V2 mechanisms tend to bridge gaps, V1 responses specifically imply continuity of contrast borders. It can be seen, then, that V2 cells show interesting new properties that one would never have guessed from their bar and edge responses• It seems that V2 is closer to contour perception than VI• It might come as a surprise that a phenomenon that gave rise to early gestalt psychology27 and that was later regarded as cognitive43 should be evident in monkey cortex• But even cats see illusory contours, as shown in behavioral studies 44'45. Thus, 'cultural differences' do not seem to be a factor. It is surprising to find these signals at a level as low as V2 (in the cat they have even been noted in area 17; Ref. 46, but see also Ref. 47), where the cells still appear to be 'driven' by the stimulus, as is the case in V1. This suggests that we are dealing with 'low-level' processes, although 'top-down' influences cannot easily be ruled out, since the recordings are from alert animals. Cognitive
have been interpreted similarly53. Physiology also predicts that adaptation with anomalous contours in one eye should affect test contours in the other eye (interocular transfer), since only cells of V2 that are nearly always binocular are involved, whereas adaptation with ordinary lines should be only partially effective in transference, since it also involves the monocular populations of V1; this has in fact been confirmed52. Within V2, anomalous-contour responses occur in the thick stripes and the interstripes, Unit 3GD5
A
27.7 It"
B
i"
.-"
•
• ':'.,. ..
" '''~-:,::
.
" ::'.':Z~'..
4.6
C
0.7 D
4.0
contours?
Cognitive theory43'4s assumes that perception results from the system trying to find the most likely explanation for given sensory data. In the case of the illusory triangle, it might consider the possibility that three sectored black discs and three angles happen to be arranged in that particular way or, alternatively, that there is a white triangle overlying three complete discs and an outline triangle - the latter is accepted as the more likely explanation. One argument against the involvement of higher centers is the relatively short latency of the anomalous-contour responses 33 (compare Refs 49, 50). Other arguments come from psychophysical experiments. Prolonged inspection of anomalous contours of one orientation results in subsequently viewed contours of near-by orientations appearing tilted, just as in the tilt after-effect with ordinary lines51.52. However, while adaptation with lines makes other lines and anomalous contours appear tilted, adaptation with anomalous contours only tilts these contours and hardly affects ordinary lines 52. This is precisely what is predicted by the physiological data, since anomalous contour-sensitive cells are only a subset of the orientation-selective cells in V2 (Refs 33, 34). The rest of these remain unadapted and thus provide unbiased orientation signals. Asymmetry found in the training effects of orientation judgement TINS, Vol. 14, No. 3, 1991
i ! I
E
0.0 2°
0.5s (2 °)
Fig. 3. The responses of a cell recorded in area V2 of the monkey visual cortex. While the monkey is fixating the cross, the cell's response field (ellipse) is traversed by a dark bar moving back and forth within a stationary light rectangle (A), by an illusory bar (B) or by a contour produced by discontinuity of lines (D). In (C), the moving notches are closed by thin (2 rain arc) lines. No stimulus is present in (E). The responses of the cell are shown on the right, where each dot corresponds to an action potential, and each row to a cycle of stimulus movement. The forward sweeps are plotted in the left halves of the display and the backward sweeps are plotted in the right halves, with a reversed time axis. Average numbers of action potentials per cycle are shown under each panel. The ceil responds to the bar (,4) as well as to the different anomalous contours (B), (19), but with the closing lines the illusory-bar response is virtually abolished (C). For perception, see Fig. 2B and C. Orientation selectivity was found to be the same for ordinary (A) and anomalous contours (B), (D). While this cell responds weakly to anomalous contours, other cells in V2 respond as strongly as to bars and edges or even more strongly. 115
A
® ®
Y
==I>
B
Fig. 4. A possible mechanism for contour perception. (A) It is assumed that contour cells sum two different kinds of signals that both originate from the same patch of retina - one for detection of contrast borders (grey), and another for detection of pattern discontinuity (light blue), The cell shown in (A) has a vertical orientation. The first type of input comes from simple or complex receptive fields of the same orientation (vertical ellipse) and the second comes from a set of end-stopped receptive fields that are distributed along this orientation, but oriented perpendicular to it (horizontal ellipses with circular disks symbolizing inhibitory end-zones; only four are shown). These cells are thought to respond to corners and line-ends, but not to edges and continual lines. Pairs of end-stopped cells with distant receptive fields are connected by non-linear elements (x) which produce a signal only if both cells are activated simultaneously. In (Z) these signals are summed, together with the first input, to produce the contour signal. (B) The mechanism is activated by the anomalous contour of a triangle and it is turned off when the open sectors are closed by circular line segments. Two corners excite a pair of end-stopped cells; the line segments inhibit them.
but not in the thin stripes of the cytochrome-oxidase pattern s4, where color-coded cells predominate sS. This might relate to the observation that illusory contours fade when the inducing shapes are displayed in equiluminant color contrast 47's6. These findings suggest that mechanisms similar to those in monkey exist in the human visual cortex. Orientation-specific adaptation and disappearance of anomalous contours at equiluminance seem to imply low-level mechanisms.
The theoretical level Physiological data p e r se are not a satisfactory explanation of the phenomenon of illusory contours 57, even though one encounters expressions such as 'physiological explanation' or 'physiological theory' in the literature. The existence of a large number of cells signaling anomalous contours (one-third of the cells of V2, which is the second largest visual area 5s) rather raises the question of why the system puts so much effort into analysing these contours - it can hardly be 116
to create an illusion. As sketched above, the detection of occluding contours is a key problem of understanding images. It seems that, as a result of evolutionary adaptation, the circuitry of V2 can detect these contours at an early stage of processing by the fast, parallel action of a large number of relatively simple modules. The principle of this process relies on the fact that various image features occur in statistical correlation with occluding contours 59, such as the abrupt changes of (1) intensity, (2) structure (see Fig. 1A), (3) binocular disparity, (4) motion and (5) the dynamic occlusion/disocclusion of the background, to list only a few possibilities of what relatively simple mechanisms could cope with. A change in disparity (3) occurs because occluding contours, by definition, are lines of discontinuity of depth. Features (4) and (5) result when the observer or the object moves. In stationary images, only (1), (2) and (3) are available; monocularly, only (1) and (2) are available. Each of these indicators taken by itself is insufficient to reveal the contours (Fig. 1B), but when combined they permit reliable detection.
A possible mechanism Figure 4 shows a simple model of a mechanism that combines evidence from contrast and discontinuity-ofbackground patterns (more sophisticated models including computer simulations exist6°'61). Contrast borders stimulate an input from classical simple or complex cells in the center (grey shading), and terminations of lines and edges excite end-stopped cells with receptive fields distributed along the contour and mainly orthogonal to it (blue shading). Pairs of these cells straddling the center field are connected by gating units to ensure that activation on one side only has no effect (so that illusory contours do not spread out from single corners). These signals are added to that of the classical input. This means that a coincidence of terminations can be equivalent to contrast (compare Ref. 62). In figures like those in Fig. 2, this mechanism is wrong, hence the illusion, but when looking at normal three-dimensional scenes, it can be largely correct. It also produces geometrical illusions when the intersected elements are not orthogonal to the contour, but are tilted 39. This is a consequence of the mainly orthogonal orientation of the end-stopped cells. Clearly, the circuits of lowlevel mechanisms must be rather simple because they have to be replicated many times over for the different locations in the visual field. Illusions are the price of simplicity; speed of parallel processing is the gain. Preattentive parallel processing Can parallel processing of anomalous contours be demonstrated in a visual search task? With ordinary lines one can detect, say, a horizontal target line among vertical lines with a reaction time that is independent of the number of lines presented 63. The same is true when bars made up of anomalous contours of the type shown in Fig. 3D are used 64. A horizontal anomalous bar pops out from an array of vertical anomalous bars. However, since the anomalous contour-sensitive cells 33'34 also respond to contrast borders, anomalous contours do not pop out when they have to compete with other contours (exclusive anomalous-contour detectors might not TINS, Vol. 14, No. 3, 1991
exist). Thus, an illusory triangle produced by a triplet of sectored discs in a group of other triplets of sectored discs cannot be detected by parallel search 65. The simplicity of the proposed mechanism (Fig. 4) should not be taken as indicating that contour perception as a whole is a simple process. As pointed out, one has to assume several different mechanisms acting in parallel, and the manner in which their outputs are combined is sophisticated. For example, binocular disparity not only generates contours 1°'3°, but also interacts with the formation of contours from pattern discontinuities:~°'66: pattern discontinuity produces depth perception by itself 67, or by interaction with motion 6~ and dynamic occlusion/disclosure69'7°. Dealing with situations of occlusion implies that there exists some representation of the hidden parts of objects as sketched in Fig. 1C ('amodal completion'aS'69), which can be demonstrated perceptually 7~. Some of these demonstrations, like those of stereoscopic illusory contours 66, were meant to support a cognitive theory. While these have disclosed much of the complexity of contour processing, it is intriguing to investigate how much cognitive reasoning can be translated into the language of low-level mechanisms ~9. It is important to note that the mechanisms present in visual cortex have evolved to a high degree of sophistication, incorporating a great deal of information about the environment, optical imaging and statistical correlations in images. This implicit knowledge 72 makes the action of these mechanisms appear to be cognitive. It could be said that perceptual hypotheses are 'wired in'. However, the resemblance is superficial. Many demonstrations show that the system does not really evaluate the hypotheses or check for consistency at a global level. Perception can be grossly inconsistent with the known geometry of space (Fig. 5). Certainly, form perception also involves processing at higher levels, and some aspects of illusory contours cannot be explained by stimulus-driven mechanisms alone 73, but the question is whether feedback to lower levels must be assumed.
Strengthening the link between physiology and perception The link with perception is indirect in all the examples discussed above, since neuronal signals and perception were compared in different subjects or even species. Perhaps the most exciting recent development is the introduction of simultaneous recording and perceptual testing. This technique was first used to compare neuronal and perceptual spectral-sensitivity thresholds in the monkey 74. By comparing the neuronal response and perceptual decision for each single-stimulus presentation, it has been shown that variations in response of certain neurons in area MT correlate with the momentary fluctuations of movement perception at threshold 75 and under conditions of binocular rivalry 76, and that microstimulation of small pools of cells hi MT can alter perception dramatically in the way predicted from the motion selectivity of these cells 77. These results demonstrate convincingly that the perception of motion direction is related to the activity in this area and might actually depend on the responses of only small groups of cells. It will be most interesting to see TINS, Vol. 14, No. 3, 1991
Fig. 5. It often seems that contour perception involves 'interpretation' of sensory information in terms of a three-dimensional layout of objects. However, this might simply reflect 'implicit knowledge' in the design of lowlevel cortical mechanisms, Global consistency is not checked, and perception can contradict known geometry of space. (Reproduced, with permission, from Ref. 28.)
whether this paradigm can be extended to more complex perceptual or cognitive functions.
Anatomical segregation of function The stripy pattern of cytochrome-oxidase activity seen in V2 has been interpreted as representing three interlaced systems for color, form and depth processing 55. This intriguing hypothesis is based on the distributions of neuronal selectivity in three sets of stripes for stimulus orientation and length, wavelength and binocular disparity, and on the pattern of connections with V1 (Refs 55, 78, 79). It has received further support from demonstrations of selective connections with the prestriate areas V4 and MT, whose specializations are in accordance with such a scheme s°-s3. However, the segregation of the selectivities in V2 is not quite as clear as might be expected from the proposed segregation of function s4'~5. In fact, this makes sense since perceptual qualities like contour, color, motion and depth are not characterized primarily by unique 'cues', but by their lawful dependence on certain stimulus conditions, including, for example, invariance with changes of contour configuration, illumination, eye movements and viewing distance. Selectivity might not reveal function. Selectivity for disparity, for example, can serve depth as well as form perception TM (see Ref. 80 for further discussion of the multiple roles of 'cues'). Cells in V4 are not more selective for wavelength than are cells in V1 or retinal ganglion cells s6, but they show color constancy4'5. Cells in V2 are not more selective for orientation than are cells in V1, but represent contours in a generalized sense 33'34. Thus, the approach of correlating physiology with perception might give a clearer picture of the courses of cortical processing than the distributions of selectivity. Recent findings in alert monkeys are encouraging. Anomalous-contour responses are found in interstripes and thick stripes, but not in thin stripes (thought to process color), while orientation 117
Acknowledgements We thank Dorothea Wemger for correcting the English, Friedrich Heitger and Lukas Rosenthaler for providing the computed contrast borders of Fig. 18, and Ernst Peterhans for modeling the background lamp. We also thank Peter O. Bishop and Peter Hammond for comments and critictsm of an earlier version of the manuscript, Supported by Swiss NF Grant 3 939 84.
selectivity is common even in thin stripes and length selectivity is equally frequent in all three sets 54. Indications of shape-from-motion processing were found to be most frequent in the thick stripes, while direction selectivity was equally rare in all three sets of stripes 4°'87. The question is, what is achieved at the higher stages of cortical processing, rather than how much of the receptive-field properties of the lower stages is preserved?
Concluding remarks Physiological studies that explicitly used perceptual phenomena for interpreting neuronal signals have been discussed. These seem to mark a promising new trend in vision research. A more representative account of the attempts to relate perception to physiology can be found in Ref. 88. In addition, the discussion of the psychophysical and theoretical studies on contour processing is limited, and thus biased by the scope of this review. Contours are necessary, even if they are subjective.
Selected references 1 Marr, D. (1980) Phil. Trans. R. Soc. London Ser. B 290, 199-218 2 Baumgartner, G. (1986) in Diseases of the Nervous System: Clinical Neurobiology (Vol. 2) (Asbury, A. K., McKann, G. M. and McDonald, W. I., eds), pp. 804-815, Saunders 3 yon der Heydt, R. (1987) Rev. Physiol. Biochem. Pharmacol. 108, 69-150 4 Zeki, S. (1983) Neuroscience 9, 741-765 5 Zeki, S. (1983) Neuroscience 9, 767-781 6 Movshon, J. A., Adelson, E. H., Gizzi, M. S. and Newsome, W. T. (1985) Exp. Brain Res. (Suppl. 11), 117-151 7 Galletti, C., Battaglini, P. P. and Fattori, P. Exp. Brain Res. (in press) 8 Sakata, H., Shibutani, H., Kawano, K. and Harrington, T. L. (1985) Vision Res. 25,453-463 9 Sakata, H., Shibutani, H. and Kawano, K. (1983) J. Neurophysiol. 49, 1364-1380 10 Komatsu, H. and Wurtz, R. H. (1988) J. Neurophysiol. 60, 580-603 11 Newsome, W. T., Wurtz, R. H. and Komatsu, H. (1988) J. Neurophysiol. 60, 604-620 12 Komatsu, H. and Wurtz, R. H, (1988) J. Neurophysiol. 60, 621-644 13 DLirsteler, M. R. and Wurtz, R. H. (1988)J. Neurophysiol. 60, 940-965 14 AIIman, J., Miezin, F. and McGuinness, E. (1985) Perception 14, 105-126 15 Tanaka, K. et al. (1986)J. NeuroscL 6, 134-144 16 Frost, B. J., Scilly, P. L. and Wong, S. C. P. (1981) Exp. Brain Res. 43, 173-185 17 Orban, G. A., Gulyas, B. and Vogels, R. (1987) J. Neurophysiol. 57, 1792-1812 18 Julesz, B. (1971) Foundations of Cyclopean Perception, University of Chicago Press 19 Poggio, G. F. (1984) in Dynamic Aspects of Neocortical Function (Edelman, G. M., Gall, W. E. and Cowan, W. M., eds), pp. 613-635, Wiley 20 Sakata, H., Shibutani, H. and Kawano, K. (1980) J. Neurophysiol. 43, 1654-1672 21 De Valois, K. K., De Valois, R. L. and Yund, E. W. (1979) J. Physiol. 291,483-505 22 Maffei, L., Morrone, C., Pirchio, M. and Sandini, G. (1979) J. Physiol. 296, 27-47 23 Hubel, D. H. and Wiesel, T. N. (1962) J. Physiol. 160, 106-154 24 Hubel, D. H. and Wiesel, T. N. (1968) J. Physiol. 195, 21 5-243 25 Marr, D. and Hildreth, E. (1980) Proc. R. Soc. London Ser. B 207, 187-217 26 Poggio, T., Gamble, E. B, and Little, J. J. (1988)Science 242, 436-440 27 Schumann, F. (1900) Z. Psychol. 23, 1-32 28 Kanizsa, G. (1979) Organization in Vision. Essays on Gestalt 118
Perception, Praeger 29 Petry, S. and Meyer, G. E. (1987) The Perception of Illusory Contours, Springer 30 Lawson, R. B. and Gulick, W. L. (1967) Vision Res. 7, 271-297 31 vonder Heydt, R., Peterhans, E. and Baumgartner, G. (1984) Science 224, 1260-1262 32 Peterhans, E., yon der Heydt, R. and Baurngartner, G. (1986) in Visual Neuroscience (Pettigrew, J. D., Sanderson, K. J. and Levick, W. R., eds), pp. 343-351, Cambridge University Press 33 yon der Heydt, R. and Peterhans, E. (1989) J. Neurosci. 9, 1731-1748 34 Peterhans, E. and vonder Heydt, R. (1989) J. Neurosd. 9, 1749--1763 35 Petry, S., Harbeck, A., Conway, J. and Levey, J. (1983) Percept. Psychophys. 34, 169--174 36 Fuld, K. and O'Donnell, K. (1984) in Sensory Experience, Adaptation, and Perception. Festschrift for Ivo Kohler (Spillmann, L. and Wooten, B. R., eds), pp. 461-469, Lawrence Erlbaum Associates 37 Spillmann, L. (1975) in Gestalttheorie der Modernen Psychologie (Ertel, S., Kemmler, L. and Stadler, M., eds), pp. 210-218, Steinkopf 38 Kennedy, J. M. (1978) Perception 7, 605-607 39 yon der Heydt, R. and Peterhans, E. (1989) in Neural Mechanisms of Visual Perception (Lam, D. M. K. and Gilbert, C. D., eds), pp. 157-170, Portfolio Publishing 40 Peterhans, E. and vonder Heydt, R. (1989) in Seeing Contour and Colour (Kulikowski, J. J., Dickinson, C. M. and Murray, I. J., eds), pp. 125-130, Pergamon 41 Hammond, P. and MacKay, D. M. (1983) J. Physiol. 337, 69-87 42 Hammond, P. and MacKay, D. M. (1985) J. Physiol. 359, 315-329 43 Gregory, R. L. (1972) Nature 238, 51-52 44 Bravo, M., Blake, R. and Morrison, 5. (1988) Vision Res. 28, 861-865 45 De Weerd, P., Vandenbussche, E., De Bruyn, B. and Orban, G. A. (1990) Behav. Brain Res. 39, 1-17 46 Redies, C., Crook, J. M. and Creutzfeldt, O. D. (1986) Exp. Brain Res. 61,469-481 47 Gregory, R. L. (1987) in The Perception of Illusory Contours (Petry, S. and Meyer, G. E., eds), pp. 81-89, Springer 48 Rock, I. and Anson, R. (1979) Perception 8, 665-4581 49 Reynolds, R. I. (1981) Perception 10, 107-115 50 Gellatly, A. R. H. (1980) Perception 9, 599-602 51 Smith, A. and Over, R. (1975) Nature 257, 581-582 52 Paradiso, M. A., Shimojo, S. and Nakayama, K. (1989) Vision Res. 29, 1205-1213 53 Vogels, R. and Orban, G. A. (1987) Vision Res. 27, 453-467 54 Peterhans, E. and von clef Heydt, R. (1989) Soc. Neurosci. Abstr. 15, 161 55 Hubel, D. H. and Livingstone, M. S. (1987) J. Neurosci. 7, 3378-3415 56 Brigner, W. L. and Gallagher, M. B. (1974) Perceptual and Motor Skills 38, 1047-1053 57 Marr, D. (1982) Vision. A Computational Investigation into the Human Representation and Processing of Visual Information, Freeman 58 Van Essen, D. C. (1985)in Cerebral Cortex (Vol. 3) (Peters, A. and Jones, E. G., eds), pp. 259-329, Plenum 59 von der Heydt, R. and Peterhans, E. (1989) in Seeing Contour and Colour (Kulikowski, J. J., Dickinson, C. M. and Murray, I. J., eds), pp. 729-734, Pergamon 60 Grossberg, S. and Mingolla, E. (1987) in The Perception of Illusory Contours (Petry, S. and Meyer, G. E., eds), pp. 116-125, Springer 61 Finkel, L. H. and Edelman, G. M. (1989) J. Neurosci. 9, 3188-3208 62 Spillmann, L., Fuld, K. and Neumeyer, C. (1984) Perception 13,513-520 63 Treisman, A. (1988) Q. J. Exp. Psychol. (A) 40, 201-237 64 Gurnsey, R. and Humphrey, G. K. (1990) Invest. Ophthalmol. Visual Sci. 31, 107 (ARVO suppl.) 65 Grabowecky, M. and Treisman, A. (1989) Invest. Ophthalmol. Visual Sci. 30, 457 (ARVO suppl.) 66 Harris, J. P. and Gregory, R. L. (1973) Perception 2,235-247 67 Coren, S. (1972) Psychol. Rev. 79, 359-367 68 Yonas, A., Craton, L. G. and Thompson, W. B. (1987) Percept. Psychophys. 41, 53-59 69 Michotte, A., Thines, G. and Crabbe, G. (1964) Les Complements Amodaux des Structures Perceptives (Studia psychoTINS, Vol. 14, No. 3, 1991
Iogica), Publications Universitaires de Louvain 70 Kaplan, G. A. (1969) PercepL Psychophys. 4, 193-198 71 Nakayama, K. and Shimojo, S. (1989)in NeuralMechanisms of Visual Perception (Lain, D. M. K. and Gilbert, C. D., eds), pp. 281-296, Portfolio Publishing 72 De Weert, C. M. M. and van Kruysbergen, N. A. W. H. (1987) in The Perception of Illusory Contours (Petry, S. and Meyer, G. E., eds), pp. 165-170, Springer 73 Bradley, D. R. and Dumais, S. T. (1975) Nature257, 582-584 74 Sperling, H. G., Crawford, M. L. J. and Espinoza, S. (1978) Mod. ProbL Ophthalmol. 19, 2-18 75 Newsome, W. T., Britten, K. H., Movshon, J. A. and Shadlen, M. (1989) in Neural Mechanisms of Visual Perception (Lam, D. M. K. and Gilbert, C. D., eds), pp. 171-198, Portfolio Publishing 76 Logothetis, N. K. and Schall, J. D. (1989) in Neural Mechanisms of Visual Perception (Lain, D. M. K. and Gilbert, C. D., eds), pp. 199-222, Portfolio Publishing 77 Salzman, C. D., Britten, K. H. and Newsome, W. T. (1990)
Invest. OphthalmoL Vision Sci. 31,238 (ARVO suppl.) 78 Livingstone, M. S. and Hubel, D. H. (1984) J. NeuroscL 4, 309-356 79 Livingstone, M. S. and Hubel, D. M. (1987) J. Neurosci. 7, 3371-3377 80 DeYoe, E. A. and Van Essen, D. C. (1988) Trends Neurosci. 11,219-226 81 Zeki, S. and Shipp, S. (1988) Nature 335, 311-317 82 Shipp, S. and Zeki, S. (1989) Eur. J. Neurosci. 1,333-354 83 Zeki, S. and Shipp, S. (1989) Eur. J. Neurosci. 1,494-506 84 DeYoe, E. A. and Van Essen,D. C. (1985) Nature 317, 58--61 85 Levitt, J. B. and Movshon, J. A. (1990) Invest. Ophthalmol. Vision Sci. 31, 89 (ARVO suppl.) 86 Schein, S. J., Marrocco, R. T. and De Monasterio, F. M. (1982) J. Neurophysiol. 47, 193-213 87 Peterhans, E. and vonder Heydt, R. (1989) InvesL Ophthalmol. Vision ScL 30, 299 88 Spillmann, L. and Werner, J. S. (1990) Visual Perception The Neurophysiological Foundations, Academic
ERRATUM In the article 'Dopaminergic innervation of the cerebral cortex: unexpected differences between rodents and primates' by B. Berger, P. Gaspar and C. Verney (January 1991, Vol. 14, pp. 21-27) an error was introduced to the colour shading of part of Fig. 1. The DA cell groups A9 and A10 shown in the section at the bottom of the Figure were given the wrong shades of blue. We reproduce the correct Figure here and apologise to the authors and readers for this error.
par
v
rs
$m
tern
c i n g ~ ent
Pf / ~ . ~ . ~ i
6
ns 1
RAT 3turn
,.--, "-~
~''
• "
MACAQUE 10ram
A10-A9
European NeuroscienceConference A meeting on K÷ channels and neuronal mechanisms will be held at Chateau de Fillerval, Mouy, France from 24-28 November, 1991. A maximum of 65 attendees, in addition to 25 invited speakers, will be selected. All expenses for participants from Eastern European countries will be met by the organizers. In addition, one or more fellowships will be offered to Eastern European participants on a competitive basis to enable them to visit a Western European laboratory involved in research on the topic of the meeting. Interested applicants should contact the organizer, Dr Y. Ben-Ari, INSERM U-29, HOpital de Port-Royal, 123 Bd de Port-Royal, 75674 Paris Cedex 14, France as soon as possible with a brief c.v., publication list and area of interest (two pages maximum). TINS, Vol. 14, No. 3, 1991
119