Functional segregation and temporal hierarchyof the visual perceptive systems K. MOUTOUSSIS AND S. ZEKI*

Wellcome Department of Cognitive Neurology, University College London, Gower Street, LondonWCIE 6BT, UK.

SUMMARY

In extending our previous work, we addressed the question of whether different visual attributes areperceived separately when they belong to different objects, rather than the same one. Using our earlierpsychophysical method, but separating the attributes to be paired in two different halves of the screen,we found that human subjects misbind the colour and the direction of motion, or the colour and theorientation of lines, because colour, form, and motion are perceived separately and at different times.The results therefore show that there is a perceptual temporal hierarchy in vision.

1. INTRODUCTION how the visual image in the brain is produced.This relatively new concept is entirely counter-intuitive, given the brain's integrative efficiencyin providing a unitary image. It supposes thatvision is an active process, with the parallel andapparently simultaneous processing of theseparate attributes being due to the fact that thekind of operation that the brain has to undertaketo construct one attribute, say form, issubstantially different from the kind of operationit has to perform to construct another attribute,say colour, a difference that requires a differentneural apparatus (Zeki 1981; Livingstone &Hubel 1988). But the parallel processing of theseparate attributes naturally raises the problem ofintegration, of how these separate attributes arebrought together in another active process to giveus our unitary image of the visual world. Thereis an irony here, again at the expense of thevisual physiologist: for it is the demonstration offunctional specialization and parallel processingthat has led us all now to seek to understand thatintegrative mechanism in the brain which,because of its high efficiency, initially inhibitedus from realizing the extent of the division oflabour that is required to construct a visualimage.

The visual brain is a remarkably efficient organ,capable of providing, within a fraction of asecond, a visual image in which all the differentattributes of the visual scene—form, colour,motion—are apparently seen in precisespatiotemporal registration. It is, perhaps, thevisual physiologist who has been the mostobvious victim of this efficiency; it led him toassume for a long time that vision is anessentially passive process, consisting of twostages—the 'impression' of an 'image' of thevisual world on the retina and its 'transmission'to the primary visual cortex (Zeki, 1993).Implicit in this terminology, commonly used byneurologists until recently, was the suppositionthat the 'reception' of the visual image by theprimary visual cortex led to 'seeing', a passiveactivity vaguely resembling a photographicprocess, an analogy that is commonly drawn.The more active process, the interpretation ofwhat is 'seen', was considered to be the functionof the surrounding, and at that time, ill-defined,'association' cortex. Such a view thereforeseparated seeing from understanding, andassigned a separate cortical seat to each. But ithad another consequence too: it inhibited aconsideration of the magnitude of the task thatthe brain has to perform to construct the visualimage, and thus the study of the equally complexstrategy and neural apparatus that it hasdeveloped to undertake this task.

In a previous paper (Moutoussis & Zeki 1997),we began by addressing the question of whetherthe attributes of the visual scene that areprocessed separately are brought into precisetemporal registration. We showed that colour andmotion, two attributes about whose separateprocessing there is no doubt (Zeki 1973, 1974;Ramachandran & Gregory 1978; Cavanagh et al.1984, 1985; Carney et al. 1987), are alsoperceived separately, with colour having a leadtime of about 50-100 ms over motion. This ledus to propose (i) that just as the processingsystems are separate and operate in parallel, so do

A hallmark of that strategy is the parallelprocessing of different attributes of the visualscene in many different, geographically separate,locations (Zeki 1978; Livingstone & Hubel1988). The discovery of that strategy, throughanatomical and physiological work during thepast 25 years, has led to a different concept of

1

the perceptual systems; and further (ii) that thenormal brain binds the outcome of the activity inits different processing systems, and that ittherefore misbinds in terms of what occurs in realtime. This misbinding is very different from theone that is dependent upon spatial location, andis compromised following lesions in brain areascritical for space perception (Friedman-Hill et al.1995). It seemed worthwhile to extend our earlierstudy for two reasons. First, we wanted to testthe generality of our proposition that the brainperceives the two attributes, colour and motion,separately. If this statement is generally true,then it should apply regardless of whether thesetwo attributes belong to the same object, as inour previous study, or not. We thereforedeveloped a new method for this study, in whichthe two attributes belong to different objects, anapproach that had the advantage of validating ourprevious method and conclusions.

Hubel & Livingstone 1985; Shipp & Zeki1985), the physiological differentiation betweenform and colour in the prestriate cortex is lesswell defined than that between motion andcolour. Moreover, while the indifference of V5cells to colour is acknowledged (Zeki 1974;Gegenfurtner et al. 1994), their indifference toform has been debated, some maintaining thatthe cells of V5 are not exigent with respect toform (Zeki 1974; Albright 1984), others findingthe contrary (Maunsell & Van Essen 1983).Whatever their exact preferences, thephysiological homogeneity of V5 in terms ofdirectional motion selectivity sits in contrast tothe apparently more heterogeneous physiology ofV4. In the latter, large concentrations of colour-and wavelength-selective cells are separated fromeach other by orientation selective cells (Zeki1975, 1983; Desimone & Ungerleider 1986;DeYoe et al. 1994), leading one to suppose thatmonkey V4 is also concerned with form(Desimone & Schein 1987) and with form inassociation with colour (Zeki 1990). Thus, theneat separation found between colour and motionis not so obvious in the colour and formsystems. Moreover, there have been persistentreports of the presence, in Vl (Ts'o & Gilbert1988) and V2 (Gegenfurtner et al. 1996), of cellsthat have dual selectivity for orientation and forcolour, but the proportion of such cells has notbeen very impressive (Livingstone & Hubel1984; Hubel & Livingstone 1987). This dualselectivity is therefore not as significant as thephysiologically more impressive segregation ofdifferent attributes at the processing level whichthis and other laboratories have reported (DeYoe& Van Essen 1985; Hubel & Livingstone 1985;Shipp & Zeki 1985). Nevertheless, it seemedinteresting to learn whether, at the perceptuallevel, there is a sufficiently precise temporalintegration to give such dual selective cells amore precise integrative role in perception. If so,one might expect form and colour to be perceivedin precise temporal registration, quite unlikecolour and motion.

We were concerned, next, to learn whether ourconclusion that the perceptual systems areseparate generalizes to attributes other thanmotion and colour and whether there is,therefore, a temporal hierarchy in vision. To doso, we added simple form to our repertoire andstudied the relative perceptual times involved inperceiving colour when paired with form, andmotion when paired with form. This gave usthree pairings: form-colour; form-motion, andcolour-motion. The choice of these pairings, asan addition to the colour—motion pairing, isdeliberate; both the form-motion and the form-colour pairs have perceptual and physiologicalambiguities in them that are far less prominent inthe colour-motion pair. While the perceptualdistinction between motion and colour has solidanatomical and physiological foundations,reflected in the distinct physiologies of areas V4and V5 and the distinct M- and P-derivedpathways leading to them (Livingstone & Hubel1988), there is no such unanimity about thephysiological and anatomical separation of formand motion or form and colour. With regard tothe latter, most would agree that every colour,being confined in space, has a form and thatevery form has a colour. Similarly, the brainconstructs colour by comparing the wavelengthcomposition of the light reflected from onesurface and that reflected from surroundingsurfaces (Land 1974). This comparison requiresthe presence of a border, and the border has aform. Hence the impossibility of separatingcolour completely from form, either perceptuallyor computationally. Physiologically, too, theseparation between form and colour and form andmotion is not as neat as that between motion andcolour. While the wavelength and directionallyselective cells in Vl and V2 are largely restrictedto their own compartments and have separatedestinations within prestriate cortex (Livingstone& Hubel 1984; DeYoe & Van Essen 1985;

2. MATERIALS AND METHODS

Essentially, the psychophysical method that wehave used here is similar to the one usedpreviously (Moutoussis & Zeki 1997), withimportant modifications. Subjects were asked toview a pattern generated by a Macintoshcomputer on a Mitsubishi Diamond Pro 17TXmonitor operating with a resolution of 640 x 480pixels at 67 Hz vertical refresh rate. In any onesitting, only two attributes (e.g. colour andmotion) were present on the screen but, unlikeour previous experiment, each attribute waspresented in one half of the TV monitor only(figure 1). The variation of each attribute can bedescribed by a square-wave oscillation of periodT = 0.537s presented at various phase differences

2

Figure 1. The four experiments done in this study as they appeared on the 30˚ x 22˚ computer screen.The screen was always split into two halves, with a different attribute change taking place in each half.The grey, red, and green colours in the motion-colour and form-colour experiments, shown in (b) and(d) were isoluminant. The 3˚ x 3˚ coloured checkerboards were stationary on the screen, the colouredsquares switching from red to green and back (X, Y, Z coordinates for red = 3.39, 7.09, 1.39; for green= 12.3, 6.73, 0.732). The oriented lines (3˚ x 0.5˚ in size) were also stationary, and switched from lefttilt to right tilt and back. The checkerboards made of grey and black squares moved either vertically(when paired with colour or orientation) or horizontally (when paired with a vertically movingcheckerboard) with a speed of 19˚ s-1. Subjects were allowed to fixate the cross on the right or to movetheir gaze freely around the screen (normally they did both), and had to pair, in terms of simultaneity,the stimuli appearing on the left half with those appearing on the right half of the screen. In (a) thevertical motion of the squares presented on the left half of the screen is compared to the horizontalmotion presented on the right half. In (b) the orientation of lines presented on the left is compared tocolour of checks on the right. In (c) the vertical motion of the squares on the left is compared to theorientation of the lines on the right. In (d) the vertical motion on the left is compared to the colour ofthe checks on the right.

with respect to the square wave oscillationdescribing the change in the other attribute; thephase differences covered the whole range of 0-

calculate the perception time differences betweenfour pairs of attributes. These were as follows.

360˚. The phase difference between the twoattributes was varied in steps of 10˚, each phasedifference being presented four times in randomorder. Each presentation lasted at most 14 s; thesubjects made their choice by selecting from twooptions using the computer's mouse during thisperiod, or after the termination of each trial.

(a) Control experiment: motion and motion(figure 1a)Two identical, moving, grey and blackcheckerboard patterns appeared on the two halvesof the screen. The pattern to the left changeddirection along the vertical axes and the one tothe right along the horizontal axes, following theoscillations described above

Subjects had to decide which values of the twoattributes were present on the screensimultaneously. The data derived from all foursubjects were averaged to obtain a value withstandard errors.This new method was used to

(b) Colour and form (figure lb)The right half of the TV monitor contained acoloured checkerboard pattern, while the left partcontained grey bars (equiluminant to both the

3

green and the red) on a black background. All thebars were tilted by 23° from the vertical to theright or to the left, following the typicaloscillation described above. The task of thesubjects was to decide what the colour of thechecks was when the bars were tilted to the right,and what colour the checks were when the barswere tilted to the left.

subjects' task in this case was to decide whichdirection of vertical motion occurred when thebars were tilted to the nght or left.

d) Colour ard motion (figure 1d)The upward and downward motion of a grey-black checkerboard pattern on the left of thescreen had to be paired with the colour of anidentical stationary checkerboard which changedfrom red-black to green-black, both colours being

(c) Motion and form (figure 1c)The left half of the screen contained a moving(up—down) grey-black checkerboard pattern,while the right half contained the oriented barspattern of the colour-form experiment. The

equiluminant to the grey. The subjects' task wasto decide what colour the checkerboard was whenthe motion on the left was upwards, and

Figure 2. A diagrammatic representation of the relationship between the change in the colour on theright half of the monitor and the motion on the left. The square-wave oscillations describing theattribute changes over time are shown top left (veridical stimulus) and top right (the perceivedstimulus). Both oscillations have the same period T and are presented at various phase differences withrespect to each other; 0˚, 90˚, 180˚, and 270˚ are shown as examples. The bottom part of the figureshows, on the left, a polar plot of the percentage of time green was present on the right half of thescreen when there was upward motion present on the left half. The polar plot on the right depicts theconsequences of a hypothetical lead of colour over motion. Here the motion oscillations have beenshifted to the right by T/4, resulting in a 45˚ anticlockwise rotation of the curve.

4

likewise, when it was moving downwards.Theresults described below were obtained from foursubjects who viewed the displays in theconfiguration described above. As a control forpositional effects, we repeated all the experimentswith three of the four subjects; in these repeats,we changed the position of the attributes on thescreen compared to the position that theyoccupied in the first experiment (for example,colour present on the left and motion on theright). The results we obtained were identical,and thus the position on the screen (right or left)does not seem to bias relative perceptual times inany way.

attributes. A rotation angle was calculated bytreating each point on the graph as a vector fromthe centre, calculating a mean vector, and finallymeasuring its angle to the vertical. From thisvector, the difference in relative perceptual timescould be calculated, since 360˚ = 537ms, andthus each degree is 1.5 ms (perceptual time isdefined as the time interval between theappearance of the stimulus on the screen and itsperception by the brain). The polar data were'unwound' and plotted as a population on aCartesian graph centred on the rotation angle. Asecond population was calculated by shifting theoriginal one so that it was centred on zero.Finally, a Z score was calculated by carrying outa Wilcoxon rank sum test (Howell 1992) on thetwo populations. This was converted to a p valueby reference to the two-tailed normaldistribution.

In the example of colour and motion, figure 2(upper left) shows four conditions in which thephase differences are arranged in such a way thatthe upward motion can be entirely correlated withgreen or red (0˚ and 180˚), or the colour canchange midway during the motion (at 90˚ and270˚). If the time difference between colour andmotion perception is equal to ∆t and colour isperceived first, then the colour present on thescreen at any time t is not perceived togetherwith the motion present on the screen at thattime but with the motion at time t—∆ t. Thisresults in shifting the motion waveform to theright with respect to the colour waveform by anamount equal to ∆t (see figure 2 upper right).This shift results in an additional phasedifference between the two oscillations equal to(∆t/T x 360˚, where T is the period of theoscillation.

3. RESULTS

The expected theoretical result if all attributes ofthe visual image are seen in precise temporalregistration, or are synchronized to 'time 0', isshown in figure 2 (lower left). Figure 2 (lowerright) shows the result that might be expected ifone of the attributes (in this case colour) is seenfirst. This would result in a counterclockwiserotation, the extent of which depends upon thedifference in time between the perception ofcolour and of motion; if motion is perceivedfirst, the rotation will be clockwise and, onceagain, dependent upon differences in perceptualtimes. The difference in relative times betweenthe perception of colour and of motion is givenby the degree of rotation: the entire 360˚ circle isequivalent to 537 ms, so that an anticlockwiserotation of, say, 45˚ would amount to a timedifference in favour of colour of about 67 ms.

For each experiment, the set of responses fromthe four subjects were summed, and plotted aspolar curves; these constitute the response curves.For these curves, the percentage of times that theanswer was 'X is the property of the right half ofthe screen when Y is the property of the left half'is plotted for each phase difference. In theexample of figure 2 (lower part) the per cent ofthe time that the green colour was present in onehalf of the screen while the upward motion waspresent in the other half (which is the same as theper cent of the time that the red colour in onehalf and the downward motion in the other werepresent together) is plotted. If there is no delaybetween the perception of colour and motion, theresponse curves should be broadly similar inshape and position to this veridical curve (figure2 lower left). If, on the other hand, there is adifference in the two perception times tested,then the resulting response curve should deviatefrom the veridical one clockwise if motion isperceived first and anticlockwise if colour isperceived first (figure 2 lower right); the greaterthe rotation from the veridical, the greater theseparation in perceptual time between the two

The average response curves of the split-screenexperiment are shown in figure 3: (a) shows theresults of the motion versus motion experiment,which is a statistically insignificant deviation (z= 0.94, p = 0.35) of 0.4 ms in the average score.There is therefore no difference in the perceptiontime between the up-down versus left-rightmotion, presumably because they are bothperceived by the same specialized perceptualsystem. A very similar result was obtained whenwe compared motion and motion with our oldermethod, where subjects had to decide thedirection of motion embedded within the squareswhen the latter were moving up or down, that is,when the two motions belonged to the sameobject. These results would seem to constitute agood perceptual demonstration of thesimultaneous perception of the two componentsof motion.

5

Figure 3. The averaged responses (shown in white) and standard deviations (shown in green) of foursubjects to (a) motion and motion, (b) colour and orientation, (~) orientation and motion, (d) colourand motion. It is clear that colour is perceived before form, which is perceived before motion.Furthermore, these differences are numerically consistent between them. No difference was foundbetween the times necessary for the perception of vertical versus horizontal motion

The colour versus orientation curve is shown in(b). Significant at the p < 0.0001 level (z =7.94), it shows a 41.9˚ anticlockwise rotation,i.e. colour is perceived before orientation by 63ms. The orientation versus motion curve, shownin (c), is rotated by 34.8˚ anticlockwise (z =5.53,p<0.0001), showing that orientation is perceived52 ms before motion. Finally, the colour versusmotion curve, shown in (d), is rotated by 79.0˚anticlockwise (z = 11.76, p < 0.0001), showingthat colour is perceived before motion by 118ms. This result is similar to the one in ourprevious paper (Moutoussis & Zeki 1997), if onetakes into account the standard errors of the tworesponse curves. The overall similarity betweenthis result and our previous one, where both thecolour and the motion of the same objectchanged, shows that 'binding' of the two

attributes to the same object does not in any wayinfluence the perception time differences betweendifferent systems.

In summary, the results from this study extendour previous conclusions, and show that theperceptual systems for the different attributes areseparate, just as are the processing systems, atleast insofar as form, colour, and motion areconcerned. This conclusion gains further validityfrom the numerical additivity of the results infigure 3: colour precedes orientation by 63ms,which in turn precedes motion by 52ms. Theresult that colour precedes motion by 118 ms isnot far from the expected 63 ms + 52 ms = 115ms. In fact, when one looks at individual results,such a close additivity is seen in only two of thefour subjects, although all saw colour before

6

form before motion. Equally, and in spite of thisuniformity in the hierarchy of perceptual times,there was a bigger difference between colour andorientation than between orientation and motionfor two subjects only: in one subject thedifference in the orientation-motion pairing isbigger than the difference in the orientation-colour pairing, while in the last the differenceswere almost equal. These individual variationsshould be expected.

us, intuitively much the most appealing solutionwas an anatomical convergence, a strategy bywhich the results of operations performed in allthe specialized visual areas would be relayed toone area or a set of areas—which would then actas the master integrator and, perhaps, evenperceptive areas. Apart from the logical difficultyof who would then perceive the image providedby the master area(s) (Zeki 1993), there is a morepractical difficulty—the failure to observe an areaor a set of areas that receive outputs from all theantecedent visual areas. Thus the convergentanatomical strategy is not the brain's chosenmethod of bringing this integration about. Infact, even when two areas such as V4 and V5project to a common third area—as they do, inboth parietal and temporal cortex—the zone ofdistribution from the two areas overlaps onlyvery minimally in the territory of the third, ananatomical fact for which we have coined theterm 'juxtaconvergence' (Shipp & Zeki 1995).Once again, this suggests that even if two areaswith different specializations were to project to athird area, the integration will be brought aboutnot by direct convergence, but by the action ofinterneurones linking the two territories.

4. DISCUSSION

The work described here is an extension of Ourprevious work which showed that, when colourand motion belong to the same object (i) colouris perceived before motion, and (ii) that thisresults in a 'misbinding' of these two attributes,leading to perception of conjunctions that departfrom the reality on the screen. With our newmethod, each attribute is presented in one-half ofthe screen only, and is continuously varied inthat half so that the two attributes no longerbelong to the same object; instead, the changesin one attribute occur in a different part of thescreen than those in another, leading to acondition in which the perception of eachattribute is spatially independent from the other.The new departure is simpler, and we shouldreally have done it first; it allows us to testperception time differences between differentattributes more conclusively because theperceptual 'binding' that subjects are asked to dohere is more clearly time-based and less object-bound. Nevertheless, if our hypothesis ofindependent perceptual systems, each with itsown perceptual time, is correct then this newmethod should give results identical to theoriginal one; this is exactly what we haveobserved.

There are of course other anatomicalopportunities for the compartments of the brainrepresenting different visual attributes to interactwith each other in an integrative manner; such aninteraction could, for example, occur in areaV2where the stripes containing cells with differentselectivities are concentrated. These stripes areconnected to one another by a rich system ofhorizontal connections (Rockland 1985; Levitt etal. 1994) that could conceivably provide theanatomical opportunity for some level ofintegration. But timing studies of V2 (and alsoV1) neurones have shown that the motion-relatedcells are the earliest, and that colour-related cellsare the last cells to be activated (Munk et al.1995; Nowak et al. 1995). This is quite theopposite hierarchy to that of the perceptualhierarchy revealed in our studies; it is thereforeunlikely that V2 can be regarded as a perceptualintegrator area where all the different visualattributes come together (Shipp & Zeki 1989;Roe & Ts'o 1995). Our perceptual studies lead usto believe that no such area is indeed necessary;even if it exists, however, it is not equipped witha time compensator mechanism that could bringthe different attributes back into their correcttemporal relation. Our hypothesis, derived fromour previous and present work, is that theperceptual systems are separate, just like theprocessing systems, and that, as far as the brainis concerned, it will process two separateattributes such as colour and motion, or motionand form, separately and perceive themseparately. Because the perceptual systems fordifferent attributes are separate, it does not matter

The results that we obtained, both here and inour previous work, are counter-intuitive althoughthey reinforce each other. They suggest that thenormal brain does not perceive the differentattributes of the visual scene at the same time,nor is it able to synchronize its differentperceptual systems to 'time 0'. Instead, the brainmis-binds in terms of real time, which is thesame thing as saying that it only synchronizesthe results of its own operations. We are notunaware of the more general significance of thesefindings, in terms of theories of perception, sinceit leads us to the general theory of a temporalhierarchy of the visual perceptive systems.

Ever since the demonstration of functionalspecialization within the primate (Zeki 1978),including human (Zeki et al. 1991) visual brain,the question of integration has imposed itself.Whatever solution the brain has adopted to dealwith this, it has turned out not to be the kind ofsolution thought of by neurobiologists. To all of

7

whether the attributes belong to the same or todifferent objects.

REFERENCES

Albright, T. D. 1984 Direction and orientationselectivity of neurons in visual area MT of themacaque. J. Neurophysiol. 52, 1106-1130.

Our results thus lead us to conclude thatintegration is brought about by the visual brainusing a strategy that is different from what seemsthe most obvious one—that of integrating theresults of its many operations in such a way thatwhat happens in real time is brought into precisetemporal registration perceptually. Instead, thebrain appears to ignore real time and synchronizewith respect to its own time. No doubt, it doesthat in an ingenious and counter-intuitive waywhich we have yet to uncover.

Carney, T., Shadlen, M. & Switkes, E. 1987Parallel processing of motion and colourinformation. Nature 328, 647--649.

Cavanagh, P., Booglin, J. & Favreau, O. E.1985 Perception of motion in equiluminouskinematograms. Perception 14, 151-162.

Cavanagh, P., Tyler, C. W. & Favreau, O. 1984Perceived velocity of moving chromatic gratings.J. Opt. Soc. Am. A 1, 893-899.

Desimone, R. & Schein, S. J. 1987 Visualproperties of neurons in area V4 of the macaque:sensitivity to stimulus form. J Neurophysiol. 57,835--868.

This work was supported by the Wellcome Trust.We thank Robin Edwards for the computerprograms and statistical analyses.

Desimone, R. & Ungerleider, L. G. 1986Multiple visual areas in the caudal superiortemporal sulcus of the macaque. J. Comp.Neurol. 248, 164 - 189.

DeYoe, E. A. & Van Essen, D. C. 1985Segregation of efferent connections and receptivefield properties in visual area 2 of the macaque.Nature 317, 58-61.

DeYoe, E. A., Felleman, D. J., Van Essen, D.C. & McClendon, E. 1991 Multiple processingstreams in occipito-temporal visual cortex.Nature 371, 151 - 154.

Friedman-Hill, S. R., Robertson, L. C. &Treisman, A. 1995 Parietal contributions tovisual feature binding: evidence from a patientwith bilateral lesions. Science 269, 853-855.

Gegenfurtner, K. R., Kiper, D. C. &Fenstemaker, S. B. 1996 Processing of color,form, and motion in macaque areaV2. VisualNeurosci. 13,161-172.

Gegenfurtner, K. R., Kiper, D. C., Beusmans, J.M. H., Carandini, M., Zaidi, Q & Movshon, .J.A. 1994 Chromatic properties of neurons inmacaque MT. Visual Neurosci. 11, 455 - 466.

Howell, D. C. 1992 Statistical methods forpsychology, 3rd edn. Belmont, CA: DuxburyPress.

Hubel, D. H. & Livingstone, M. S. 1985Complex-unoriented cells in a subregion ofprimate area 18. Nature 315, 325-327.

8

Hubel, D. H. & Livingstone, M. S. 1987Segregation of form, color, and stereopsis inprimate area 18. J Neurosci 7, 3378--3415.

Shipp, S. D. & Zeki, S. M. 1989 Theorganization of connections between areas V5 andV2 in macaque monkey visual cortex. Eur. JNeurosci. 1, 333-354.

Land, E. H. 1974 The retinex theory of colourvision. Proc. R. Inst. 49, 23-58. Shipp, S. & Zeki, S. 1995 Segregation and

convergence of specialised pathways in macaquemonkey visual cortex. J. Anat. 187, 547-562.Levitt, J. B., Yoshioka, T. & Lund, J. 1994

Intrinsic cortical connections in macaque visualareaV2: evidence for interaction between differentfunctional streams. J. Comp. Neurol. 342, 551-570.

Ts'o, D. Y. & Gilbert, C. D. 1988 Theorganization of chromatic and spatial interactionsin the primate striate cortex. J Neurosci. 8, 1712-1727.

Livingstone, M. S. & Hubel, D. H. 1984Anatomy and physiology of a color system inthe primate visual cortex. J Neurosci. 4, 309 -356.

Zeki, S. M. 1973 Colour coding in rhesusmonkey prestriate cortex. Brain Res. 53, 422-427.

Livingstone, M. S. & Hubel, D. H. 1988Segregation of form, color, movement, anddepth: anatomy, physiology, and perceptionScience 240, 740-749.

Zeki, S. M. 1974 Functional organization of avisual area in the posterior bank of the superiortemporal sulcus of the rhesus monkey. J Physiol.(Lond.) 236, 549—573.

Maunsell, J. H. & Van Essen, D. C. 1983Functional properties of neurons in middletemporal visual area of the macaque monkey. I.Selectivity for stimulus direction, speed, andorientation. J Neurophysiol 49,1127-1147.

Zeki, S. M. 1975 The functional organization ofprojections from striate to prestriate visual cortexin the rhesus monkey. Cold Spring HarborSymp. Quant. Biol. 40, 591-600.

Zeki, S. M. 1978 Functional specialization inthe visual cortex of the monkey. Nature 274,423-428.

Moutoussis, K. & Zeki, S. 1997 A directdemonstration of perceptual asynchrony invision. Proc. R. Soc. Lond. B 264 393-399.

Zeki, S. 1981 The mapping of visual functionsin the cerebral cortex. In Brain mechanisms ofsensation (ed. Y. Katsuki, R. Norgren & S.Masayasu), pp. 105-128. NewYork.

Munk, M. H.J., Nowak, L. G., Girard, P.,Chounlamountri, N. & Bullier, J. 1995 Visuallatencies in cytochrome oxidase bands ofmacaque area V2. Proc. Natn. Acad. Sci. USA92, 988-992. Zeki, S. 1983 The distribution of wavelength

and orientation selective cells in different areas ofmonkey visual cortex. Proc. R. Soc. Lond. B217, 449-470.

Nowak, L. G., Munk, M. H.J., Girard, P. &Bullier,J. 1995 Visual latencies in areas Vl andV2 of the macaque monkey. Visual Neurosci. 12,371 - 384. Zeki, S. 1990 A century of cerebral

achromatopsia. Brain 113 1721 -1777.Ramachandran, V. S. & Gregory, R. L. 1978Does colour provide an input to human motionperception? Nature 275, 55-56.

Zeki, S. 1993 A vision of the brain Oxford, UK:Blackwell.

Rockland, K. S. 1985 A reticular pattern ofintrinsic connections in primate area V2 (area18). J. Comp. Neurol. 235, 467-478.

Zeki, S. M., Watson, J. D. G., Lueck, C. J.,Friston, K. J., Kennard, C. & Frackowiak, R. S.J. 1991 A direct demonstration of functionalspecialization in human visual cortex. JNeurosci. 11, 641-649.Roe, A W. & Ts'o, D. Y. 1995 Visual

topography in primate V2: multiplerepresentation across functional stripes.JNeurosci. 15, 3689 - 3715.

Shipp, S. D. & Zeki, S. M. 1985 Segregation ofpathways leading from area V2 to areas V4 andV5 of macaque visual cortex. Nature 315, 322-325. Received 21 March 1997; accepted 1 May

1997

9

10