200 Research Paper

Human colour discrimination based on anon-parvocellular pathwayTom Troscianko*, Jules Davidoff†, Glyn Humphreys‡, Theodor Landis§, Manfred Fahle¶, Mark Greenlee#, Peter Brugger§ and William Phillips¥

Background: Traditionally, colour information is assumed to be carried by neuralchannels in the parvocellular pathway and to be encoded in an opponentmanner, while other, non-parvocellular, spectrally non-opponent channels arethought to play no part in colour vision. But is the parvocellular pathway the onlyway that colours can be discriminated in human vision? We studied two patientswith cerebral achromatopsia, who lack conscious colour perception but arenevertheless able to make use of colour information. In particular, weinvestigated whether, in these patients, colour discrimination is mediated by theparvocellular pathway.

Results: The achromatopsic patients carried out a forced-choice colour- andluminance-discrimination task, and showed clear evidence of unconscious colourprocessing, consistent with previous studies. We added different types ofluminance noise to see when this unconscious colour information could bemasked. The results of the colour-discrimination-with-noise and the brightness-non-additivity experiments showed a double-dissociation between patients. Thisindicates that, in one patient, unconscious colour discrimination may besubserved by a spectrally non-opponent mechanism, which does not have thecharacteristics of the parvocellular pathway and which is responsive to fastflicker. Spectral sensitivity, contrast sensitivity and motion perceptionexperiments confirmed that this patient lacks a working opponent parvocellularsystem. The second achromatopsic patient showed evidence of a residualparvocellular system.

Conclusions: Our results show that chromatic discrimination need not bemediated by neural mechanisms, the parvocellular system in particular, normallyassumed to subserve conscious colour perception. Such discrimination may bemediated by a neural subsystem which responds to fast flicker, is spectrallynon-opponent, and supports normal motion perception.

BackgroundThe neural basis of colour vision appears to be well-estab-lished. There are three cone types in the retina, and thesignals produced by these are encoded so that the visualsystem obtains opponent signals — red–green (R–G) andyellow–blue (Y–B), where Y = R + G, which signal thechromatic response of the system, and a further signal(R + G) that gives the visual system’s luminance (achro-matic) response [1,2]. How does this view relate to what isknown about post-receptoral visual pathways?

These visual pathways are known to be divided into twodistinct types, ‘fast’ and ‘slow’ [3], which recent terminol-ogy refers to as the magnocellular (‘M’) and parvocellular(‘P’) pathways, respectively [4]. The magnocellular systemis commonly regarded as being colour-blind, and beinginvolved primarily in the perception of certain kinds ofmotion [5]. It is thought to receive summed inputs fromthe R and G cones (and so to be spectrally non-opponent),

and is responsible for the photopic spectral sensitivityfunction, V(l), from which luminance is defined [6–9].This pathway receives no input from B cones.

The parvocellular system mediates both colour vision,being the site of the chromatic opponent responsesdescribed above, and high-acuity luminance vision[2,5,7,10]. This pathway receives inputs from all threecone types, and so is a suitable carrier of chromatic infor-mation. There can be no doubt that, without this pathway,we could not have the rich experience of colour vision; butit may not be the only route by which chromatic informa-tion is transmitted. There is physiological evidence sug-gesting that the magnocellular system can detect bordersbetween isoluminant colours (colours of equal luminance)[5,11,12], and evidence is emerging that there is a furthersystem, known as the koniocellular (K) system, whichsends projections to the primary visual cortex, but whosefunctional characteristics are as yet unknown [13]. To

Addresses: *Perceptual Systems Research Centre,Department of Psychology, University of Bristol, 8 Woodland Road, Bristol BS8 1TN, UK.†Department of Psychology, University of Essex,Colchester CO4 3SQ, UK. ‡School of Psychology,University of Birmingham, Edgbaston, BirminghamB15 2TT, UK. §Department of Neurology, UniversityHospital Zürich, Haldenbach D, Frauenklinik 26,CH 8091 Zürich, Switzerland. ¶Sektion VisuelleSensorik, University Eye Hospital, Waldhörnlestr 22,D-72072 Tübingen, Germany. #NeurologischeKlinik mit Abt. für Neurophysiologie, Hansastr 9, D-79104 Freiburg i. Br., Germany. ¥Department ofPsychology, University of Stirling, Stirling FK9 4LA,Scotland, UK.

Correspondence to: Tom TrosciankoE-mail address: Tom.Troscianko@Bris.ac.uk

Received: 18 October 1995Revised: 21 December 1995Accepted: 21 December 1995

© Current Biology Ltd ISSN 0960-9822

Current Biology 1996, Vol 6 No 2:200–210

Research Paper Non-parvocellular colour discrimination Troscianko et al. 201

what extent do such non-parvocellular systems contributeto the discrimination of colour in humans? It is possible,even likely, that if non-parvocellular systems do con-tribute to chromatic discrimination, they may not generatea conscious percept of colour; any non-parvocellular chro-matic effect must therefore be studied using performancecriteria rather than subjective experience.

Our question, then, is whether the human visual systemcan perform colour discrimination without using thespecific comparisons known to exist in the parvocellular,opponent system. We have examined this issue by study-ing patients with cortically-based impairments of colourvision, as damage to the primary colour pathway mightreveal the function of an alternative, almost certainlyweaker, pathway. We tested two subjects with cerebralachromatopsia [14]; patients with this condition reportseeing the world in ‘shades of grey’ but can, paradoxically,detect borders between isoluminant colours without beingable to name the colours [15–17]. It is therefore likely thatthere is some residual colour vision in such patients, buttheir main neural pathway either is completely function-less and the information is being processed through a sec-ondary system, or the main pathway may not be providinga clear output to whichever system subserves the con-scious perception of colour. In either case, it is of interestto see how the residual, unconscious, colour vision works.Evidence from the patient of Heywood et al. [16,18] sug-gests that a residual parvocellular opponent systemresponse may be responsible.

Our first experiment was designed to establish whetherthe two patients could process colour information. Thesubjects had to make a ‘same’ or ‘different’ judgementabout two parts of an image, spatially separated on adisplay monitor. The parts of the image differed in lumi-nance, colour, or colour and luminance together, allowingus to test whether the subjects’ discrimination perfor-mance improved when there was a colour difference aswell as a luminance (intensity) difference. Such animprovement would be indicative of the (unconscious)use of colour information. To investigate which neuralchannels were responsible for any such improvement inperformance, luminance noise could be added to thedisplay; the temporal frequency of this noise could bemade to affect primarily the parvocellular or magnocellularsystems.

This experiment yielded clear evidence of unconsciouscolour processing in both subjects, but suggested that onesubject was using a residual parvocellular system, whereasthe other was using a non-parvocellular system. A bright-ness non-additivity test [18–20] was also carried out onboth subjects, in which the brightnesses of red and greenfields were matched to a yellow field; the reds and greenslook brighter than a yellow at the same luminance, and this

is assumed to be a particular feature of the parvocellular,opponent system. We predicted normal (non-additive)results for one subject, but perfectly additive results for theother subject. This prediction was upheld, sustaining thehypothesis that the latter subject had no functioning parvo-cellular system that could process colour information.

Several further experiments were performed on this lattersubject. One test examined his spectral sensitivity func-tion (the amount of energy needed to detect a monochro-matic test spot on a white or dark background), and theresults were again consistent with a loss of the parvocellu-lar system. We also measured his spatio-temporal contrastsensitivity (the amount of contrast required to detect agrating whose luminance varies sinusoidally as a functionof space and time). We predicted that a loss of the parvo-cellular system should be evident in a particularly markedloss of high spatial and low temporal frequencies. Thepredicted pattern of results was observed. Finally, wetested his motion perception [21], which established thatthis subject has relatively undamaged non-parvocellular,motion-detecting systems.

We interpret these results as indicating that a non-parvo-cellular system may be used to process colour informationin human vision. Such a system is additional to the mainparvocellular system and does not contribute to colourperception; rather, it can use colour information to obtain astronger signal in a non-colour task such as brightness dis-crimination. Colour information may therefore modulatethe perception of other attributes in an unconsciousmanner.

ResultsRed–green colour- and luminance-discrimination taskwith added noiseIn this experiment, the two achromatopsic subjects, HJAand WM (see Materials and methods) were presentedwith a display in which the top and bottom parts were thesame, or differed in colour, luminance, or both. Their taskwas to say which of two images (presented sequentially intime) contained a difference between the two halves ofthe display. They were not required to say what the dif-ference was, just when it occurred. To prevent the taskbeing possible simply by noticing a local discontinuity inthe image, a white buffer strip was added between thetwo halves of the display in most tests [16]. Luminancenoise was added to the display in some tests; the additionof noise makes the task harder, by forcing the system tointegrate across the image before a correct response canbe computed. However, such noise can only make thetask harder if it can be encoded by the particular systemwhich is responsible for that subject’s chromatic discrimi-nation. Dynamic (25 Hz) noise cannot be encoded by theparvocellular system, whereas static noise is optimallyencoded by this system and only poorly encoded by the

magnocellular system. Figure 1 shows examples of thestimuli used in this experiment (which is discussed indetail in Materials and methods).

The two subjects were tested with a variety of stimuli; dif-ferences between the ‘monochrome’ (green–green) and‘red–green’ curves in the graphs shown in Figure 2 indi-cate the extent to which colour information was used byeach subject. A subject with truly monochrome perceptionwould give identical results in the two cases (this was veri-fied using a subject who was a congenital dichromat; datanot shown). In general, both patients’ monochrome andred–green curves were not coincident, indicating thatchromatic information was available to their visual systemsin some form. HJA’s performance on red–green discrimi-nation was poor with static noise and improved withdynamic noise, whereas WM showed the opposite trend.In the white-buffer, no-noise condition (in which the topand bottom fields of the stimulus were uniform), HJAgave results similar to his results with dynamic noise; forWM this condition gave results similar to the ones withstatic noise. The similarity between the no-noise and the25 Hz-noise conditions for HJA implies the action of a par-vocellular system as, in both these cases, this ‘sees’ aclean, noise-free image. For WM, on the other hand, thesimilarity lies between the no-noise and static-noise con-ditions, which both look fairly clean to a fast, non-parvo-cellular system. Both subjects performed poorly givennoisy monochrome stimuli, although somewhat betterthan chance, whereas an age-matched control subject gavesharper tuning curves and was making virtually no errorsby 20 % contrast.

These results suggest that HJA may be somewhat similar tothe patient of Heywood et al. [16,18], using a residual parvo-cellular system to distinguish between the colours, whereasWM may be using a non-parvocellular system. The inclu-sion of a bright white buffer strip between the two hemi-fields made discrimination impossible at isoluminancewhen there was added noise; without such a buffer stripboth subjects were above-chance for all ‘with-noise’ lumi-nance contrasts in the red–green conditions. However, HJAcould obtain above-chance results in the no-noise, white-buffer condition, whereas the subject of Heywood et al. [16]could not. Note also that the ‘isoluminance’ point of subjectWM did not coincide exactly with that predicted by flickerphotometry, but was occasionally shifted 10 % along thecontrast axis. However, this particular red–green contrastlevel produced data at chance on two subsequent occasions,and shifts of static isoluminance have been observed inmany studies on normal subjects (possibly because of thedependence of the match point on spatial frequency).

Were display artefacts responsible for these results? Itcould be that subjects may not detect colour modulationbut do detect small differences in the perceived contrast

of the noise in the red and green fields. We discount thisexplanation for three reasons. Firstly, there was a con-trast value at which WM’s performance was at chance(see Fig. 2b). It remained so on two further occasions.Secondly, we tested a congenital dichromat with thesame conditions, and obtained chance performance over

202 Current Biology 1996, Vol 6 No 2

Figure 1

Examples of the stimuli used in the red–green colour and luminancediscrimination task. All examples shown include a white buffer strip andstatic noise. The subject’s task was to identify the target stimuluswhich was randomly presented either before or after the non-targetstimulus. The differences between target and non-target stimuli couldbe of colour, or luminance or both. (a) A red–green target stimulus. (b)A red–green non-target stimulus. (c) A monochrome target stimulus.(d) A monochrome non-target stimulus.

a 10 % range of contrasts about the isoluminant point.Finally, the similarity of the ‘no-noise’ data and the data

where noise had little effect on performance (dynamicfor HJA, static for WM) suggest that discrimination was

Research Paper Non-parvocellular colour discrimination Troscianko et al. 203

Figure 2

Results of discrimination experiments for bothpatients under four conditions: (a) static(0 Hz) noise and no buffer between the twofields; (b) static noise and a bright white bufferbetween the two fields; (c) dynamic (25 Hz)noise and a bright white buffer between thetwo fields; and (d) no noise and a bright whitebuffer between the two fields. In all cases, thetask was to identify the temporal interval inwhich the top of the stimulus was differentfrom the bottom. The chance performancelevel is 50 %. The monochrome curves refer tostimuli differing only in the luminance of topand bottom halves. The red–green curvesrepresent exactly the same luminanceinformation as in the monochrome stimuli, butwith the addition of a red–green chromaticdifference. The ordinate in each graph is theproportion of errors made; error bars representstandard deviation about the mean.

not occurring as a result of the noise microstructure ofeach stimulus pattern.

Brightness additivity taskWe hypothesized that WM was basing his improved per-formance in the red–green tests on a neural subsystem thatwas not the parvocellular system, as 25 Hz noise abolishedthe chromatic contribution (a frequency to which the par-vocellular system is insensitive). In order to increase confi-dence in this hypothesis, we required a second dissociationbetween WM and HJA in a different test. In the bright-ness-additivity test, the subject is required to match theperceived brightness of a monochromatic yellow referencefield to red, green and yellow test fields. Normal subjectssee the red and green stimuli as being considerablybrighter than a yellow field of the same luminance. Thiseffect is known as brightness non-additivity [19,20], and isassumed to be a direct consequence of the spectral oppo-nency of the parvocellular pathway [20]. We predicted thatthe results for HJA should be the same as those of thenormal control subjects (non-additive), whereas those forWM should show perfect additivity.

As shown in Figure 3, the results for HJA were non-addi-tive and in the normal range (comparable to the fourcontrol subjects), whereas WM displayed perfect additiv-ity — his visual system based its perceived brightnesspurely on the luminance of light of different wavelengths.This is not a characteristic of the opponent parvocellularsystem, and strongly implies the action of a non-opponentsystem. Note that this result cannot arise as a consequenceof inhibition of the parvocellular system by the magnocel-lular system as there was no energy at high temporal-fre-quencies in this static stimulus. This experiment cannotprove that the magnocellular system in WM was perform-ing chromatic discrimination, as no such discriminationwas required here, but tends to exclude the possibility of astrong parvocellular involvement in his vision. However, itis occasionally found that some subjects with normalvision show additive data (J.D. Mollon, private communi-cation), and further experiments were required to rule outthe action of a parvocellular system in WM.

Spectral sensitivity testWe tested the spectral sensitivity function of WM by mea-suring the amount of energy needed to detect a monochro-matic test spot on either a light or a dark background.King-Smith and Carden [22] originally established that, innormal subjects, a light-adapting field gave a function withthree peaks around 450, 525 and 600 nm, whereas a dark-adapting field gave a single peak around 550 nm. A three-peaked function is assumed to indicate the action of aspectrally opponent parvocellular system, whereas a single-peaked function is obtained from any spectrally non-oppo-nent system. Note that the patient of Heywood et al. [16]produced a spectral sensitivity curve with three peaks,

exactly as expected if that patient had a functioning oppo-nent (P) system. As shown in Figure 4, the results of thistest on WM all showed a single peak at around 550 nm (theV(l) function), whether on a dark- or a light-adapting field.We conclude that WM did not use opponent channels ineither condition. Two normal subjects were tested as acontrol, and they showed a normal set of results (threepeaks on a light-adapting field; one peak on a dark field).

Spatio-temporal contrast sensitivity testFigure 5 shows the results obtained in the determinationof contrast sensitivity for WM and an age-matched controlsubject. We predicted a reduced loss compared to a normalage-matched control at low spatial and high temporal fre-quencies, which are characteristic of the frequencies thatare not encoded by the parvocellular system. Compared tothe control subject, WM had least sensitivity loss at thehighest temporal frequency and two lowest spatial fre-quencies tested (10 Hz, 0.25 and 0.5 c/deg). This, togetherwith his poor suprathreshold discrimination (Weber frac-tions of around 40 %), was in keeping with a lack of aparvocellular system and a paucity of frequency-tunedchannels, which are numerous in the parvocellular system.

Motion perception testIn order to determine whether WM had a reasonably well-functioning pathway other than the parvocellular pathway,we carried out a ‘missing-fundamental’ apparent-motionstimulus test (based on that described in [21]). As the

204 Current Biology 1996, Vol 6 No 2

Figure 3

Results of experiment to measure the extent of brightness non-additivity in the two patients and four normal control subjects. The‘additivity ratio’ shown on the ordinate is about 0.4–0.5 for normalsubjects and HJA. WM has an additivity ratio of unity, implying that thebrightness of red, green and yellow stimuli is accurately predicted bytheir luminance. HJA’s results are similar to those of four controlsubjects with normal colour vision. Error bars represent standarddeviation about the mean.

magnocellular pathway is assumed to be involved in theprocessing of motion and feeds to visual area 5 (V5), whichis assumed to be the equivalent of area MT in monkeys[23], we would expect a less-damaged pathway of thiskind to result in essentially normal perception of motion.This particular test presents subjects with a ‘missing-fundamental’ compound grating which can appear tomove in one of two directions depending on the inter-stimulus interval between successive frames [21]. It istherefore very sensitive to any abnormalities in motionperception, as it puts two forms of motion sensing (featureand energy based models) into opposition and yields avery precise balance point between the two.

The ordinate of the graph shown in Figure 6 shows thepercentage of ‘feature’ responses (where the subjects werereporting the motion which would be expected if theywere basing the percept on the motion of features (fixed-polarity edges) in the grating stimulus. The abscissa showsthe interstimulus interval (ISI) between successive framesof the apparent motion stimulus. The control subject (AL)and WM showed a changeover between mostly ‘energy’and ‘feature’ motion at ISIs of about 40 milliseconds. Wetherefore conclude that WM has essentially normal featureand energy motion detecting systems. Subjectively, WM

Research Paper Non-parvocellular colour discrimination Troscianko et al. 205

Figure 5

Ratios of spatiotemporal contrast sensitivity of WM and an age-matched control subject with no known clinical abnormality. The dataare plotted such that the shorter the excursion of the vertical bar belowzero, the less pronounced the loss at that spatiotemporal frequencycompared to the control subject. Note that WM’s loss is least severeat low spatial and high temporal frequencies.

CSpat/CScon

2

1

00.25 0.5 1 2 4

Spatial frequency (cpd)

1Hz 5 Hz 10 Hz

Figure 4

Log

sens

itivi

tyLo

g se

nsiti

vity

Log

sens

itivi

tyLo

g se

nsiti

vity

Log

sens

itivi

tyLo

g se

nsiti

vity

−

−

−

−

−

−

−

−

210

200

190

180

170

160

150

140400 500 600 700

400 500 600 700

400 500 600 700

400 500 600 700

400 500 600 700 800

400 500 600 700 800

−

−

−

−

−

−

−

240

220

200

180

160

140

120

−

−

−

−

−

−

−

−

160

150

140

130

120

110

100

90

Wavelength (nm)

Wavelength (nm)

Wavelength (nm)

Wavelength (nm)

Wavelength (nm)

Wavelength (nm)

−

−

−

−

−

110

100

90

80

70

−

−

−

−

−

−

140

120

100

80

60

40

−−−−−−−

70605040302010

01020

MF dark background

MF lightbackground

TT darkbackground

TT lightbackground

WM dark background

WM lightbackground

Results of spectral sensitivity experiments for the detection of anincremental 2 deg (approximately) test spot with a 500 msec duration,presented on a light (19 cd m–2) or dark (0.3 cd m–2) background.Parvo opponent coding predicts three peaks around 450, 525 and600 nm; non-opponent coding predicts a single peak around 555 nm.

Normal subjects are expected to show three peaks on a light adaptingbackground and a single peak on a dark background. WM’s functionalways had a single peak, implying a lack of opponent processing.Data from the two control subjects with normal colour visionconformed to normal expectations.

reported that the motion in these stimuli was clear andeasy to see.

DiscussionHJA’s chromatic discrimination appears to be mediated bya system that is insensitive to 25 Hz flicker. The mostlikely candidate is a residual parvocellular system which isfragile and therefore strongly affected by the presence ofstatic noise, to which it is very sensitive. The results of theadditivity experiment suggest that HJA has a normal par-vocellular colour system. This, and the fact that perfor-mance is unimpaired by the addition of 25 Hz noise,suggests that the chromatic discrimination is achieved bythis system. Such a result is broadly similar to the resultsof Heywood et al. [16,18], although there are some differ-ences (in particular, HJA performs above chance in the no-noise condition at isoluminance, whereas the subject ofHeywood et al. [16] drops to chance in this condition).

WM appears to discriminate colours using a system whichis sensitive to 25 Hz flicker. This makes it unlikely to bethe opponent parvocellular system or the non-opponentparvocellular system (which is also insensitive to this tem-poral frequency). Although it is not possible to prove thathis chromatic discrimination is mediated by a non-oppo-nent system such as magno, we made a series of predic-tions about expected results in other visual tasks whichwere sensitive to the existence (or otherwise) of an oppo-nent parvocellular system.

The five experiments described above demonstrate thatWM lacks a working opponent parvocellular system, andthat he appears to have a normal non-opponent fastsystem which subserves the perception of motion andflicker and is tuned to relatively low spatial and high tem-poral frequencies. It is therefore tempting to assume thatWM bases his unconscious chromatic discrimination onthis non-opponent magnocellular (or other fast) system.We can rule out artefactual explanations of the chromatic-discrimination task, such as ‘structure-from-wavelength’ inthe noise field, because for both patients, the results of theno-noise condition closely mirrored the results when‘ineffectual’ noise was added to the display. The ineffec-tual noise for HJA was dynamic noise whose temporalfrequency was virtually undetectable by the parvocellularsystem, which we assume to mediate his performance.The ineffectual noise for WM was static noise, which is arelatively poor stimulus for the fast non-opponent system.However, we stress that the conclusion that WM bases hischromatic discrimination on a non-opponent fast systemremains indirect and therefore tentative.

Taken together, our experiments suggest that WM has avery different neural subsystem to HJA for his covert chro-matic discrimination. Therefore, there is more than onekind of cerebral achromatopsia, with HJA and other achro-matopsic subjects (particularly MS, the subject tested byHeywood et al. [16,18]) relying on a residual parvocellularpathway. WM, while sharing many similarities with HJA(particularly achromatopsia, agnosia and prosopagnosia),appears to have a very different set of channels in his earlyvisual system. His case appears to challenge the idea thatchromatic discrimination, whether conscious or uncon-scious, must arise from the action of an opponent parvo-cellular system.

What is a non-opponent system that may be capable ofchromatic discrimination? In any system in which thereare units tuned to several ranges of a particular variable(such as wavelength), the response of any one unit isintrinsically ambiguous. For example, a given change incone response may arise from a change in wavelength orelse from a change in intensity. It is therefore necessary tocompare the responses of a population of tuned units (inthis case, cones) to establish the value of the perceptualattribute. Opponency is a way of achieving this, as thequantum catches in the different cone types are com-pared. However, the particular comparisons used in colouropponency (red–green, yellow–blue) are not the only pos-sible way to extract this information. Different mecha-nisms are assumed to exist for other attributes. Anexample here is orientation. While the output of a singleorientation-tuned unit must also be ambiguous (depen-dent on both orientation and contrast), it is often assumedthat the centroid of a distribution of responses from unitstuned to different orientations can give a measure of

206 Current Biology 1996, Vol 6 No 2

Figure 6

The percentage of reported ‘feature motion’ in a missing-fundamentalmotion stimulus test based on that of Georgeson and Shackleton [21].The abscissa of the graph shows the interstimulus interval (ISI) betweensuccessive frames of the apparent-motion stimulus. The transition from‘energy’ to ‘feature’ motion at ISIs around 40 msec is very similar forboth WM and the control subject, AL, as is the general shape of thefunction. The implication is that WM’s motion processing is normal.

WM

AL

% Reportedfeature motion

100

75

50

25

00.01 0.1 1 10 100 1000

Inter-stimulus interval (msec)

central tendency which lacks the above ambiguity [24].The definition of a centroid of three units (such as thethree cone types) would be:

c = (rR + gG + bB)/(r + g + b)

where r,g,b are the magnitudes of response in each unittype, and R,G,B are the weights given to the three units(for example, relating to position on the wavelength axis).The measure c is a value which could subserve chromaticdiscrimination without requiring any opponency as definednormally. We are not suggesting that the above equationaccounts for WM’s performance, but rather that a non-opponent determination of colour is possible in principle.

If the parvocellular system is not mediating WM’s residualcolour discrimination, what alternative system might beinvolved? An obvious candidate is the magnocellularsystem. Evidence is emerging that this can respond to iso-luminant edges [11]. It certainly has the right kind of tem-poral-frequency tuning to be effectively masked by the25 Hz noise stimuli used in this study. But, while there issome prima facie evidence for arguing that WM’s magnocel-lular system may be responsible for his colour blindsight,we would suggest caution. Quite apart from the difficultyin making a conceptual leap from psychophysical data tophysiological models, there are aspects of these resultswhich do not fit what is known about chromatic responsefrom the magnocellular system. Our chromatic discrimina-tion test used stimuli which contained different luminanceswithin each of the two parts of the screen to be compared,and a bright white buffer zone between the two halves ofthe screen. The findings of Lee et al. [11] suggest that themagnocellular system can respond to the presence of ared–green border, even at isoluminance. However, there isno evidence to suggest that this system can integrate colouracross many (luminance) borders, which is a primaryrequirement of any ‘colour’ system such as the one weassume to be acting in the chromatic discrimination test.So, while we cannot specifically exclude the possibility thatthe magnocellular system is responsible for our findings,neither can we argue that this is supported by any single-unit data. Furthermore, the results of the ‘monochrome’conditions of the chromatic discrimination test show thatboth subjects performed poorly with these stimuli (worsethan a control subject). This implies that WM does nothave a normal magnocellular or parvocellular pathway.

Are there any alternatives? One intriguing possibility lies inthe action of a third system, the K (or koniocellular) systemdescribed by Cassagrande [13]. This projects to the colourblobs in visual area 1 (V1), but is not thought to be respon-sible primarily for colour vision. Rather, Cassagrande arguesthat it may have a neuromodulatory function. It appears tobe involved in eye-movement control and may thereforehave sensitivity to rather higher temporal frequencies than

the parvocellular system. Unfortunately, there are fewphysiological data on the workings of this system. Wemention it here partly to counteract the tendency of mostresearchers to concentrate purely on the magno-parvocellu-lar dichotomy while ignoring a third, presumably impor-tant, set of neurons. A measure of its importance may bethat its number of constituent units is roughly the same asthe number of magnocellular units in the macaque monkey(about 10 % of retinal ganglion cells).

We therefore conclude that our findings, while not provid-ing unambiguous evidence for the involvement of a partic-ular non-parvocellular system colour discrimination, doimply that such a second system may be involved. Thereis an intriguing possibility that this may be the K system,but this would have to be supported by physiological data.

Another study has shown that, in subjects with normalvision, 25 Hz luminance noise can affect the chromaticcontribution to the detection of a stimulus. Troscianko[25] showed that both the detectability of motion and theduration of the motion aftereffect are affected by the pres-ence of such noise. So, there is some support in favour ofsimilar mechanisms in normal, as well as achromatopsic,visual systems. Their function is revealed by comparingperformance with monochrome stimuli against perfor-mance with stimuli which have added colour modulationbut the same luminance as the monochrome stimuli. Notethat such a comparison cannot be achieved using purelyisoluminant stimuli.

What are the implications of these results for understand-ing normal vision? Consider a real-world scene containingobjects of different colours. Typically, each object-bound-ary will have an associated luminance change (if it did not,monochrome photography could not work). There is there-fore a co-variation between colour and luminance in thevisual environment. The present results suggest that thevisual system contains units which respond to such co-vari-ation while signaling a percept of brightness. In otherwords, colour information may potentiate the action ofunits which normally encode brightness, and the consciouspercept is of brightness. Psychophysical evidence for suchcolour-luminance conjunction units is emerging [25,26].

Potential candidates for such units are cells which canlearn that luminance changes are accompanied by colourchanges. A possible model of such cells is provided byclassical conditioning, or Hebbian learning [27]. Here, acell is responsible for mediating the presence of theunconditioned stimulus — in this case, brightness.However, a conditioning stimulus (colour), which is nor-mally associated with the unconditioned stimulus, willpotentiate the cell’s activity. The cell has ‘learned’ aboutthe co-variation of colour and luminance. Such an explana-tion is extremely speculative, but may be testable by

Research Paper Non-parvocellular colour discrimination Troscianko et al. 207

208 Current Biology 1996, Vol 6 No 2

adaptation techniques such as contingent aftereffects.Some preliminary data supporting the possibility that WMhas a visual system which adapts to the coincidencebetween luminance and colour were reported recently[28]. A more detailed theoretical treatment of similar ideascan be found in literature on neural networks [29,30].

What did the stimuli look like to our subjects? They wereunable to name any of the (highly saturated) colours onthe screen. They did, however, remark that the red–greenstimuli looked different from the green–green (mono-chrome) stimuli. They tended to describe the differenceas enhanced luminance contrast, which is in keeping withthe explanation above. Beyond this, they found it hard todescribe the differences. There is an apparent similarityhere with the achromatopsic painter described by Sacks[31]. Sacks says of him that he preferred to watch black-and-white TV rather than colour TV because “he thoughtthat the tonal values of ‘decoloured’ colour TV seemeddifferent, less ‘normal’, than those of a ‘pure’ black-and-white set.” Our subjects may have been trying to saysomething similar.

The present work suggests that the human visual systemmay contain a greater diversity of routes for chromaticinformation than was assumed previously. The experi-mentation is, of necessity, indirect. However, the datahere begin to provide an answer to the following questionsposed by Julesz [32] in his list of unresolved questions invisual perception: “Is color utilised by the magnocellularsystem... and can such neurophysiological problems bedecided by psychological methods?” Our tentativeanswers are that colour is used by a non-parvocellularsystem (although we cannot be sure which), and that thepsychological methods used here do seem to provide atool for answering such questions.

ConclusionsOur results show that one of the two achromatopsic sub-jects tested may base his unconscious chromatic discrimi-nation on a neural mechanism which responds to fastflicker, is spectrally non-opponent, and supports normalmotion perception. This mechanism is therefore unlikelyto be the normal parvocellular channel, which is generallyassumed to be responsible for colour vision. The othersubject appeared to be using the residual function of anormal parvocellular opponent mechanism. Chromaticinformation may therefore be encoded by several dis-tinctly different neural channels.

Materials and methodsSubjectsHJA is male, aged 70, and had a bilateral stroke of the occipital cortex10 years previously. His case is well documented [17,33]. He showsvisual object agnosia (inability to identify common objects) andprosopagnosia (inability to recognise faces). In spite of his inability tobe conscious of seeing colour, HJA performed well on tests of colour

vision such as the Ishihara pseudoisochromatic plates and showed acolour component in the visual evoked potential [17]. This suggeststhat HJA’s colour vision might be similar to, but weaker than, that of‘normal’ subjects. WM is male, aged 74, and had a bilateral stroke ofthe occipital cortex about 8 years previously. His visual object agnosiaand achromatopsia are similar to HJA’s but he performs at chance oncolour-vision tests. His condition is also well documented in the litera-ture [34,35]. We can exclude the possibility that he is a dichromat ashis anomaloscope matches do not exhibit attenuation of the brightnessof either a red or a green field. Furthermore, his previous history —screening for his job as an electrician and screening in the Swiss army— make the probability of dichromacy very low.

Red–green colour and luminance discrimination task with added noiseWe tested both patients’ colour discrimination using a novel computertest. A monitor screen contained two fields which could differ in colour(red or green), or luminance, or both. The monitor was a Digivision 14-inch (35 cm), model no CD14 3112 H3. Frame rate was 50 Hz (non-interlaced). Half the screen was blanked off with black card as there isan artefact in most video screens by which the luminance of a givencolour decreases if its total area is increased; this results from imper-fect EHT control of the guns. The software inserted ‘dummy’ areas ofred and green in the invisible portion of the screen to keep the totalarea of each colour equal across stimuli; this was verified with aMinolta Spot Chroma meter, model CS-100. Stimuli were generatedon a Pluto II system (Electronic Graphics) controlled by a PC with 8-bitcontrol over the luminance from each gun, allowing 256 levels of red,green and blue. The mean luminance of the reference half of the screenwas 3.5 cd m–2, measured with the Minolta meter. The CIE chromaticitycoordinates (x,y) of the red and green phosphors were (0.61,0.34) red,and (0.32,0.58) green (measured with the same meter). Viewing dis-tance was 1 m. The subtense of the visible portion of the screen was7.5 deg (h) by 12.5 deg (v). Subjects saw two screens (separated intime), one of which contained a difference between the top and bottomhalves (in colour, luminance or both) and the other did not (both halveswere the same). Each screen was presented for 1 sec, with a 1 secinter-stimulus interval. Auditory warning of each screen was given, aswas auditory feedback about correct/wrong response after each trial.Between stimuli, the screen consisted of a jumbled set of block pixelsrandomly taken from all parts of the target stimulus (and therefore hadthe same mean colour and luminance as the stimuli, but no useful infor-mation for the task). Each patient’s isoluminant point between the redand green fields was measured using flicker photometry and was thenbracketed in small steps in case it did not represent the exact point atwhich the two colours had equal effective luminance.

Heywood et al. [16] required their patient to select the odd one ofthree square patches separated vertically. The oddity consisted of a dif-ference in hue or brightness. Their patient could perform above-chancewhen presented with two contiguous colours, but was at chance whenan achromatic buffer separated the two fields. Such an intense localdiscontinuity in colour and luminance must disrupt weak edge signalsof both the magnocellular and parvocellular systems; the task thusrequired the visual system to pool information from the stimulus fieldsrather than just the line where they met. We made an analogous dis-tinction by studying performance with and without a white buffer: abright white horizontal dividing bar between the top and bottom partsof the stimulus, having a width of 0.7 deg and a luminance of 85 cd m–2

(much higher than other parts of the stimulus).

Different temporal frequencies of luminance noise were added to theimage in a way described below, similar to that used by Heywood et al.[18]. The luminance contrast of the noise was always 50 % when it waspresent (this refers to the contrast between the light and dark patchesof noise in each part of the display). The mean luminance of each fieldwas controlled by the computer. 25 Hz dynamic noise does not getencoded by the parvocellular system, whereas 0 Hz (static) noise (at50 % luminance contrast) provides an input to both magnocellular and

Research Paper Non-parvocellular colour discrimination Troscianko et al. 209

parvocellular systems (but is particularly preferred by the parvocellularsystem). This is strongly supported by physiological data [5,36]. So, ifeither system were contributing to colour discrimination, we wouldexpect it to be ‘jammed’ by its optimal noise stimulus in the same wayas a weak radio signal is masked by a strong signal occupying thesame transmission channel. Dynamic noise would mask colour visionmediated by the magnocellular system, or any other system tuned tohigh temporal frequencies. Static noise would impair a weak chromaticresponse in both systems.

Dynamic luminance noise was added to the display by creating tworepresentations of the stimulus in the Pluto memory. These representa-tions had the same overall spatial structure but the block pixels (0.4deg square) of the stimulus had, in the case of luminance noise, a lumi-nance of either L + l or L – l (both cases randomly selected andequiprobable) where L was the luminance necessary to produce agiven contrast against the constant reference field and l was set so thatthe contrast between the light and dark blocks was 50 % (if noise waspresent). The value of l was zero in the no-noise condition. The Plutodisplay window was panned in the blanking interval between frames toone or other stimulus representation, in alternation; so that the nextframe, 20 msec later, would show the same block pixel with its noiseluminance reversed. Thus, each block pixel’s luminance was refreshedat 25 Hz (square-wave modulation) from L – l to L + l. The displayappeared to ‘shimmer’ at this frequency; the percept was not unlikelooking at a scene through a snowstorm of selectable contrast. Thetemporal frequency content of the dynamic noise was therefore all at25 Hz and multiples thereof; there was no energy below this value, andtherefore nothing detectable by the parvo-opponent system.

Static noise involved no refreshing of the block pixels’ luminancevalues. The noise was therefore present throughout the display time ofeach stimulus, and looked like a portion of a random-dot stereogram.Figure 1 shows an example of a monochrome and a red–green stimu-lus with the bright white buffer. The static luminance noise can be seenin each hemifield.

Red–green stimuliThe non-target stimulus was green both on top and bottom, and theluminances of the top and bottom were equal. Each half of the stimuluscontained the appropriate level and type of luminance noise (static,dynamic or none) but there was no overall difference between the twohalves. If there was a white buffer strip, such a presentation can becalled G–W–G (green–white–green). Without a buffer strip, it is G–G.

The target stimulus had either the top or bottom part coloured red, andthe other part was green. There was therefore a colour contrastbetween the two halves. The position of the red part (top or bottom)was randomly decided. In the case of isoluminant stimuli (correspond-ing to a value of zero on the luminance axes in Figure 2) there was nooverall luminance contrast between the red and green fields. For non-zero luminance contrasts the luminance of the green reference fieldwas either decreased (negative values on abscissa) or increased (posi-tive values) while the luminance of the red field remained constant. Inthese cases, there was luminance and colour contrast between the twoparts (top and bottom) of the display. These stimuli may be describedas R–W–G or G–W–R in the presence of a white buffer, and R/G orG–R in the absence of the buffer. Each red and green part containedthe appropriate type of local luminance noise (static, dynamic, or none).To a subject without colour vision (achromatopsic or dichromatic) thered and green noise fields appeared identical in the isoluminant condi-tion; this was achieved by careful gamma correction of each gun of themonitor and was verified with both types of subject.

Monochrome stimuliThe non-target stimulus was exactly the same as a non-target stimulusin the R–G experiment: a screen in which both top and bottom partswere green, without any overall luminance difference between them, butwith identical noise of the appropriate type. These stimuli are therefore

G–W–G or G–G depending on whether the buffer strip was present ornot. The target stimulus was again G–W/G or G–G but now one half ofthe stimulus (equiprobably top or bottom) had a higher or lower lumi-nance than the constant reference half. This was true for all stimuliexcept those with a luminance contrast of zero in Figure 2. These werecatch stimuli in which there was no difference at all between target andnon-target stimuli — all contained identical fields and chance perfor-mance was expected. This was basically a test of the degree of ran-domness of the computer presentation: a strongly non-random stimulusselection would allow the subject to learn to perform at above-chancelevels without a visual stimulus. This was never observed in the data.For all stimuli with non-zero contrast, their top and bottom parts had dif-ferent luminance. They can be characterised as G–G or G–W–Gstimuli.

Subject’s taskEach subject was told that he would see two stimuli, each preceded bya warning tone from the computer. He was told that one stimulus wouldcontain no difference between the top and bottom parts of the screen,whereas in the other stimulus (the ‘target’) there would be an overall dif-ference between the two halves of the screen. This difference could beof luminance alone, or of colour, or of both. However, the achromatopsicsubjects perceived only differences of brightness between the twohalves of the screen. Their task was to say which stimulus was thetarget. In all trials, the subject had to press a button labelled ‘1’ or ‘2’according to which stimulus (first or second) was the target. The taskwas therefore a two-interval, forced-choice, method of constant stimuli.Auditory feedback was given after each trial to indicate whether theresponse had been correct or not. The subject was not asked to reportthe nature of the perceived difference (and, indeed, found it hard to doso). Sufficient practice was given to ensure that each subject showedgood (above-chance) performance in at least some conditions. Pilotexperiments were carried out to find a range of stimulus contrast valueswhich was sufficiently broad and yet did not extend the testing period tounmanageable durations. Breaks were given whenever required — thecomputer program would pause if no response was made within 10 secof the second interval being shown. All conditions were presented inrandomised order (a process controlled by the computer). Stimuli werepresented in blocks of ten pairs, with each condition being a cumulativesum of ten such blocks. The dependent variable (% correct score) wasthus based on ten measures, each measure obtained from one block often stimulus pairs. All the data for conditions with the white buffer andnoise (static or dynamic) were obtained in a single testing session. Dataon other conditions (no buffer, and white buffer without noise) wereobtained on separate occasions.

Brightness additivity testThis experiment used a simple technique which relies on the fact thatmonochromatic red and green lights appear brighter than monochro-matic yellow lights of the same luminance. We used a standard anom-aloscope (Nagel in the case of WM; Pickford-Nicholson in the case ofHJA). Each subject was presented with a split-field stimulus, consistingof a monochromatic red field on the left and a monochromatic yellowone on the right, and asked to equate the perceived brightness of thetwo fields. Next, the left stimulus was replaced with a monochromaticgreen stimulus of the same luminance, and the subject was againasked to make a heterochromatic brightness match. Finally, the leftstimulus was set to an equal mixture of the red and green stimuli, whichlooked yellow to a normal observer. The subject was again asked toequate the brightnesses of the two hemifields. If brightness additivitywere perfect, then we would expect the last luminance to be the sameas the arithmetic mean of the previous two. The extent of the failure ofthis process of additivity can be expressed as a ratio of the actualchosen luminance in the last condition to the luminance which wouldshow additivity. A number of less than unity indicates a failure of additiv-ity. These measures were taken five times each and a mean obtained.Four normal subjects were tested as controls. Two were authors of thispaper (TL and TT), and two were undergraduates naive to the purposesof the experiment. All four had normal colour vision, as determined by

anomaloscope matches. All gave non-additivity values in line withreported data in the classic literature on the topic.

Spectral sensitivity testThis experiment measured the spectral sensitivity functions of WM andtwo normal controls, on light- and dark-adapting fields. The stimulusparameters were chosen to be as similar as possible to those used byHeywood et al. [16]. Subjects were shown a 2 deg (approximately) testspot with a duration of 500 msec, on either a light (19 cd m–2) or a dark(0.3 cd m–2) background. A method of limits was used to determine theincrement threshold for each test wavelength used. The wavelengthbandwidth of each test spot was approximately 10 nm. Each value rep-resents the mean of three repetitions. The prediction was that WM’scurve would have a single peak at about 550 nm (the same shape asthe V(l) function) in both conditions. Normal subjects show threepeaks in the light-adapted condition and one peak in the dark-adaptedcondition [22]. Two authors of this paper (MF and TT) served ascontrol subjects.

Spatio-temporal contrast sensitivity testWe tested the spatio-temporal contrast sensitivity function of WM’svisual system, using a VSG system (Cambridge Research Systems)and an adaptive procedure for the measurement of thresholds. Con-trast thresholds and discrimination thresholds were obtained at severalspatial and temporal frequencies. Control data were obtained from anage-matched subject.

Motion perception testGeorgeson and Shackleton [21] described a stimulus which consistedof a ‘missing-fundamental’ grating which is presented in an intermittentway such that, for certain interstimulus intervals (ISIs), ‘feature’ motion isseen in one direction, and ‘energy’ motion is seen in the opposite direc-tion. (Here, feature refers to edges of a fixed polarity, and energy to theFourier composition of the display). For normal subjects, a crossoverpoint occurs at which no overall motion is seen. This represents thecrossover point between seeing feature and energy motion, alsoreferred to as long-range and short-range motion. A normal crossoverpoint together with the appropriate percept on either side of it is indica-tive of a normally-functioning motion system. We tested subject WMusing a stimulus which was very similar to the one used by Georgesonand Shackleton, in order to see whether his motion perception showeda crossover at the same ISI as an age-matched control subject. Thestimulus was a missing-fundamental grating presented on a Tektronix608 oscilloscope with a green (P31) phosphor and a frame rate of100 Hz, and a luminance of approximately 28 cd m–2. The period of thegrating stimulus was 1 c/deg. The stimulus was presented in apparent-motion with different ISIs and the subject had to respond to whether thestimulus was seen to be moving left or right (forced-choice).

Acknowledgements We thank Gavin Brelstaff for the programming of the colour/luminance task;Steve Hammett for the programming of the motion perception test; AnyaHurlbert, Alan Cowey, Claire Prince and John Mollon for helpful criticism;Anne Kurtenbach and Lukas Rüttiger for the loan of the spectral-sensitivityequipment and calibration software, and the two subjects for their patienceand good humour in lengthy testing sessions. The investigation was sup-ported by grant number D/ER1/9/4/2034/102/RARDE from the DefenceResearch Agency.

References1. Boynton RM: Human colour vision. New York: Holt, Rinehart and Winston;

1979.2. DeValois RL, DeValois KK: A multi-stage color model. Vision Res 1993,

33:1053–1065.3. Enroth-Cugell C, Robson JG: The contrast sensitivity of retinal ganglion

cells of the cat. J Physiol 1966, 187:517–552.4. Hubel DH, Livingstone MS: Segregation of form, color, and stereopsis

in primate area 18. J Neurosci 1987, 7:3378–3415.5. Schiller PH, Logothetis NK: The color-opponent and broad-band chan-

nels of the primate visual system. Trends Neurosci 1990, 13:392–398.

6. DeValois RL, DeValois KK: Neural coding of color. In Handbook of Per-ception V. Edited by Carterette EC, Friedman MP. New York: AcademicPress; 1975:117–166.

7. Ingling CR, Tsou BH-P: Spectral sensitivity for flicker and acuity criteria.J Opt Soc Am [A] 1988, 5:1374–1378.

8. Vos JJ, Walraven PL: On the derivation of the foveal receptor primaries.Vision Res 1971, 11:799–818.

9. Smith VC, Pokorny J: Spectral sensitivity of foveal cone photopigmentsbetween 400 and 500 nm. Vision Res 1975, 15:161–171.

10. Ingling CR, Martinez E: The spatiochromatic signal of the R-G channel.In Colour Vision. Edited by Mollon JD, Sharpe LT. London: AcademicPress; 1983:433-444.

11. Lee BB, Martin PR, Valberg A: Nonlinear summation of M- and L-coneinputs to phasic retinal ganglion cells of the macaque. J Neurosci 1989, 9:3378–3415.

12. Saito H, Tanaka K, Isono H, Yasuda M, Mikami A: Directionally selectiveresponse of cells in the middle temporal area (MT) of the macaquemonkey to the movement of equiluminous opponent colour stimuli.Exp Brain Res 1989, 75:1–14.

13. Cassagrande VA: A third parallel visual pathway to primate area V1.Trends Neurosci 1994, 17:305–310.

14. Meadows JC: Disturbed perception of colours associated withlocalised cerebral lesions. Brain 1974, 97:615–632.

15. Mollon JD, Newcombe F, Polden PG, Ratcliff G: On the presence of threecone mechanisms in a case of total achromatopsia. In Colour VisionDeficiencies. Edited by Verriest G. Bristol: Adam Hilger; 1980:130–135.

16. Heywood CA, Cowey A, Newcombe F: Chromatic discrimination in acortically colour blind observer. Eur J Neurosci 1991, 3:802–812.

17. Humphreys GW, Troscianko T, Riddoch MJ, Boucart M, Donnelly N,Harding GF: Covert processing in different visual recognition systems.In The Neuropsychology of Consciousness. Edited by Milner AD, RuggMD. London: Academic Press; 1992:39–68.

18. Heywood CA, Cowey A, Newcombe F: On the role of parvocellular (P)and magnocellular (M) pathways in cerebral achromatopsia. Brain1994, 117:245–254.

19. Tessier M, Blottiau F: Variations des caracteristiques photometrics del’oeil aux luminances photopiques. Revue d’Optique 1951, 30:309–322.

20. Guth SL, Lodge HR: Heterochromatic additivity, foveal spectral sensitiv-ity, and a new color model. J Opt Soc Am 1973, 63:450–462.

21. Georgeson MA, Shackleton TM: Monocular motion sensing, binocularmotion perception. Vision Res 1989, 29:1511–1523.

22. King-Smith PE, Carden D: Luminance and opponent-color contributionsto visual detection and adaptation and to temporal and spatial integra-tion. J Opt Soc Am 1976, 66:709–717.

23. Zeki S, Watson JDG, Frackowiak RSJ: Going beyond the informationgiven: the relation of illusory visual motion to brain activity. Proc R SocLond [Biol] 1993, 252:215–222.

24. DeValois RL, DeValois KK: Spatial Vision. Oxford: Oxford University Press;1990.

25. Troscianko T: Contribution of colour to the motion aftereffect andmotion perception. Perception 1994, 23:1221–1231.

26. Gur M, Akri V: Isoluminant stimuli may not expose the full contributionof color to visual functioning: spatial contrast sensitivity measure-ments indicate interaction between color and luminance processing.Vision Res 1992, 32:1253–1262.

27. Hebb DO: The Organisation of Behaviour. New York: Wiley-Interscience;1949.

28. Troscianko T, Prince C, Fahle M, Regard M: The uses of high-temporal-frequency chromatic information in visual perception. Perception 1995,24S:59.

29. Phillips WA, Kay J, Smyth D: The discovery of structure by multi-streamnetworks of local processors with contextual guidance. Network1995a; 6:1–22.

30. Phillips WA, Kay J, Smyth D: How local cortical processors that max-imise coherent variation could lay foundations for representationproper. In Neural Computation and Psychology. Edited by Smith LS,Hancock PJB. Berlin, Springer Verlag; 1995b:117–136.

31. Sacks O: An Anthropologist on Mars. London: Picador; 1995.32. Julesz B: Some strategic questions in visual perception. In Representa-

tions of Vision. Edited by Gorea A, Fregnac Y, Kapoula Z, Findlay J. Cam-bridge: Cambridge University Press; 1991:331–349.

33. Humphreys GW, Riddoch MJ: To see but not to see. Hove: LawrenceErlbaum; 1987.

34. Grüsser O-J, Landis T: Visual agnosias and other disturbances of visualperception and cognition. In Vision and Visual Dysfunction Vol 12: VisualAgnosias. Edited by Cronly-Dillon J. London: Macmillan; 1991:209;394–395.

35. Scheidler W, Landis T, Rentschler I, Regard M, Baumgartner G: A patternrecognition approach to visual agnosia. Clin Vision Sci 1992, 7:175–193.

36. Schiller PH, Logothetis NK, Charles ER: Function of the colour-opponentand broad-band channels of the visual system. Nature 1990,343:68–70.

210 Current Biology 1996, Vol 6 No 2