Discussion Post 3-4 paragraphs --Visual Perception Disorders

Colour agnosia impairs the recognition of natural but not of non-natural scenes

Tanja C. W. Nijboer, Maarten J. van der Smagt, Martine J. E. van Zandvoort, and Edward H. F. de Haan

Universiteit Utrecht, Utrecht, The Netherlands

Scene recognition can be enhanced by appropriate colour information, yet the level of visual processing at which colour exerts its effects is still unclear. It has been suggested that colour supports low-level sensory processing, while others have claimed that colour information aids semantic categorization and recognition of objects and scenes. We investigated the effect of colour on scene recognition in a case of colour agnosia, M.A.H. In a scene identification task, participants had to name images of natural or non-natural scenes in six different formats. Irrespective of scene format, M.A.H. was much slower on the natural than on the non-natural scenes. As expected, neither M.A.H. nor control participants showed any difference in performance for the non-natural scenes. However, for the natural scenes, appropriate colour facilitated scene recognition in control participants (i.e., shorter reaction times), whereas M.A.H.’s performance did not differ across formats. Our data thus support the hypothesis that the effect of colour occurs at the level of learned associations.

INTRODUCTION

What is the function of colour vision? Subjectively, colour greatly enriches visual experience, but more importantly, colour plays an important role in pattern detection and object recognition, as colour is a very prominent property of objects. As a result, knowledge of object colour might help to recognize objects better, along with luminance boundaries, shading, texture, and contours (e.g., Livingstone & Hubel, 1987).

Colour information has been shown to improve object recognition. Coloured images of objects are recognized faster than achromatic images of the same objects (Biederman & Ju, 1988; Ostergaard & Davidoff, 1985),1 but this only holds for certain types of objects (Humphrey, Goodale, Jakobson, & Servos, 1994; Tanaka & Presnell, 1999)— namely, objects that are usually strongly associated with a specific colour—that is, for which the colour is “diagnostic” (Tanaka & Presnell, 1999). Interestingly, colour has no influence on object

1 Note, however, that Biederman and Ju (1988) also showed that using a mask eliminated this facilitatory effect of colour.

Correspondence should be addressed to Tanja Nijboer, Utrecht University, Department of Experimental Psychology,

Helmholtz Institute, Heidelberglaan 2, NL-3584 CS Utrecht, The Netherlands (E-mail: [email protected]).

Tanja Nijboer was supported by NWO Grant 051.02.080 of the Cognition Program of the Netherlands Organization for

Scientific Research (NWO). Maarten van der Smagt was supported by the NWO Innovational Research Grant 451.02.094

(VENI scheme) of the Netherlands Organization for Scientific Research (NWO). We would like to thank M.A.H for his generous

participation. We thank Jennifer Steeves and colleagues for making their elaborate stimulus set available to us. Also, we thank Ryota

Kanai for useful discussion of many aspects of this article.

152 # 2007 Psychology Press, an imprint of the Taylor and Francis Group, an informa business http://www.psypress.com/cogneuropsychology DOI:10.1080/02643290600989541

COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (2), 152 – 161

verification (Biederman & Ju, 1988; Ostergaard & Davidoff, 1985). Apparently, appropriate colour facilitates finding the correct name of an object in memory, whereas it does not influence the com- parison of a given name with a shown object. Ostergaard and Davidoff (1985) suggested that objects are listed in semantic representation as a col- lection of physical attributes. One of these attri- butes is specific colour information. They claim that colour being an attribute in the lexicon makes it possible to prime linked object names in the same lexicon. This explains why colour facili- tates object naming, but not object verification.

Recognition of objects high in colour diagnos- ticity (usually natural objects) is more affected by colour manipulations (e.g., incongruently coloured images) than recognition of objects low in colour diagnosticity (usually non-natural, man-made objects; Joseph & Proffitt, 1996; Price & Humphreys, 1989). When colour is not informative, object recognition appears to be based on other cues, such as shape and texture cues (Oliva & Schyns, 1997; Tanaka & Presnell, 1999).

Neuropsychological studies have shown that some brain-injured patients with impaired object recog- nition abilities demonstrate improved performance when identifying coloured images of objects compared to incongruently coloured or achromatic images (Humphrey et al., 1994; Mapelli & Behrmann, 1997). Humphrey et al. (1994) showed that colour facilitated object recognition in D.F., a patient with visual form agnosia. This patient has a profound deficit in object recognition on the basis of shape, size, and orientation, but a relatively spared ability to recognize objects on the basis of colour and texture (Humphrey et al., 1994; Milner & Heywood, 1989; Milner et al., 1991). Colour specifically improved recognition of natural objects for both D.F. and control participants. From the studies mentioned above, it can be concluded that appropriate colour facilitates the recognition of the objects high in colour diagnosticity, whereas these objects are recognized more slowly and less accu- rately when inappropriately coloured.

Results found in object recognition studies can be extended to recognition of more complex stimuli, like scenes (Oliva & Schyns, 1997; Steeves et al., 2004).

A scene is usually defined as a semantically coherent, nameable view of an environment, composed of multiple discrete objects. In natural scenes (e.g., canyon, seashore, forest, desert), there is a more con- sistent colour pattern that serves as a recognition cue than there is in non-natural scenes (e.g., city, road, shop, room). Again, scenes high in colour diagnosti- city (natural scenes) are more affected by colour mani- pulations than are scenes low in colour diagnosticity (non-natural scenes; Oliva & Schyns, 1997; Steeves et al., 2004). Moreover, Steeves et al. (2004) showed that the form agnosic patient D.F. used colour information to categorize scenes.

It is still under debate at which level in the visual pathway the beneficial effects of colour for recognizing natural scenes and objects occurs. Cases have been made for lower, sensory levels (e.g., Wurm, Legge, Isenberg, & Luebker, 1993) as well as for higher, cognitive processing stages (e.g., Humphrey et al., 1994). As mentioned above, Humphrey et al. (1994) concluded that colour facilitates naming of natural objects when presented in their diagnostic colour. Moreover, they concluded that this facilitation occurs at a high (-er) level of visual analysis, where knowledge of object properties, such as colour, is stored.

In contrast, Wurm et al. (1993) concluded that for recognizing objects, colour information plays a role at a sensory level. They investigated the role of colour in object recognition from the perspective that colour differences, like luminance differences, may provide reliable information for scene seg- mentation. For ecological reasons, they used images of food throughout their experiments. Their hypothesis was that colour may be particu- larly useful, when either the stimulus or the sensory processing is degraded, as in the case of low-pass filtered images, or low vision, respect- ively. In their first two experiments, they showed that colour speeds up object recognition about equally for normal and degraded vision (for both low-pass filtered images and low vision). To investigate the origin of the effect of colour, they carried out two more experiments. They presented partici- pants with images of a same object from different perspectives—that is, the prototypical view and three views that deviated systematically from the

COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (2) 153

EFFECT OF COLOUR AGNOSIA ON SCENE RECOGNITION

prototypical view. The results demonstrated that the effect of colour increases when images deviate more from the prototypical view. Wurm et al. (1993) suggested that nonprototypical views of objects are more difficult to recognize, and there- fore benefit more from colour, because recognition is then “more dependent on surface character- istics”. Their final experiment was directly aimed at the question of whether explicit object-colour knowledge plays an additional role in the advan- tage of colour in object recognition. Wurm et al. (1993) used information theory to calculate the informational value of participants’ colour knowl- edge for a fixed set of food items, based on ratings from the same participants to what degree colour would be helpful for recognition. The algorithm showed that the recognition of, for example, a lemon or a carrot, would benefit more from colour information, than, for example, an apple or a pepper. Subsequently, Wurm et al. (1993) observed that this object characteristic, which they termed “colour diagnosticity”, did not corre- late with the colour advantage found in the algori- thm. They argued that explicit knowledge about food colour does not account for the advantage of colour in real-time object recognition. Their overall conclusion is that the primary role of colour in object recognition is sensory in nature.

In the present study we have the unique opportu- nity to further investigate the suggested low-level sensory colour processing versus the higher cognitive level of colour processing by examining the role of colour information in scene recognition in a patient with colour agnosia. Patients with colour agnosia have intact colour perception, as demonstrated by normal performance on, for instance, hue-matching tasks, but have no explicit colour knowledge (e.g., Nijboer, van Zandvoort, & de Haan, 2006). If scene recognition is improved by specific knowledge about which colours are appropriate for specific scenes, then we expect to find a reduced effect of colour on scene recognition in patients with colour agnosia. If instead, the colour diagnosticity effect is derived from low-level processes, agnosic patients should, similar to control participants, benefit from appropriate colour information. In other words, by using a scene recognition task with a colour

agnosic patient, we can distinguish between the above-mentioned higher and lower level expla- nations of the effect of colour in visual recognition.

CASE STUDY

M.A.H. is a 44-year-old male, who came to our attention after an infarct in the right cerebellar hemi- sphere, as confirmed with a magnetic resonance imaging (MRI) scan (see Figure 1). A more detailed description is presented in van Zandvoort, Nijboer, and de Haan (in press), and only a short summary is given here. Neuropsychological assessment in the acute phase showed, apart from residual dizziness, some mild deficits in memory functioning. At follow-up, two years later, his performance on all neuropsychological tasks tapping the major cognitive domains of reasoning, language, visual perception and construction, verbal memory, visual memory, executive functioning, and neglect was above average, taken into account his age and education (see Table 1, for neuropsychological data).

Surprisingly, he showed an absolute inability to perform the Token Test (de Renzi & Faglioni, 1978). He could not match the spoken colour name with coloured tokens, and on sub- sequent testing he was unable to name the primary colours presented to him on a monitor. Additionally, his ability to perceive colours was evaluated on three different tests. His performance on the Ishihara test for colour blindness (Ishihara, 1977) and the Farnsworth – Munsell 100 hue test (Farnsworth, 1957) was well in the normal range. In addition, an experimental task was used in which the threshold of the just-noticeable difference was established for three primary visual sensory cues (shape, colour, luminance, apparent motion; de Haan, Heywood, Young, Edelstyn, & Newcombe, 1995). M.A.H. was shown three simultaneously presented stimuli. Two stimuli in each test were identical, and the third differed in shape, colour, luminance, or apparent motion. Task difficulty changed in a 3-down, 1-up manner. M.A.H. encountered no problems in any of these tasks: He outperformed the control participants on the luminance test,

154 COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (2)

NIJBOER ET AL.

and he was within normal range on the shape, colour, and apparent motion tests (see Table 1).

Despite good colour perception, M.A.H. turned out to be impaired in categorizing and naming colour on the basis of hue. He performs at chance level on a task where he has to decide whether an object is depicted in the veridical colour or not (e.g., blue tomato). He claims that he has always had this problem. In daily life, he uses strategies to infer the colour of objects—for example, comparing the surface properties of an object with that of an object with a known colour (e.g., his shirt or grass), or he uses bright- ness cues (he is very proficient in judging bright- ness), and light colours are often identified as yellow or pink and dark ones as red or blue.

Influence of luminance and colour inversions in scene recognition

Our experiments were based on the experimental design of Steeves et al. (2004). The most important

conditions in their experiment for examining colour diagnosticity were normally coloured and inversely coloured images. Intrinsic to this colour manipulation is the fact that luminance infor- mation is inverted as well. This luminance inver- sion might complicate the recognition of scenes and might thus partly explain the poorer recog- nition of inversely coloured images. Since colour agnosic patients are perfectly able to use luminance information, it was essential to rule out this explanation beforehand. Therefore we tested 16 control participants on the scene recognition task to which a new format was added—namely, inverted greyscale.

Method

Participants A total of 16 control participants (12 student con- trols and 4 age-matched controls; mean age ¼ 29.9 years; SD ¼ 8.1) and M.A.H. (44 years)

Figure 1. MRI (T2SE) for M.A.H. showing no brain abnormalities in the occipito-temporal cortices.

COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (2) 155

EFFECT OF COLOUR AGNOSIA ON SCENE RECOGNITION

participated in the experiment. All participants had normal or corrected-to-normal vision and reported that they were not colour blind, which was confirmed with the Ishihara test for colour vision (Ishihara, 1977).

Apparatus and stimuli Images were generated by a PC and were pre- sented on a laCie Electron 22-inch blue IV display. Screen resolution was 1,024 � 768 pixels, and the computer was running in 256- colour mode. The viewing distance was 80 cm.

The stimuli were adapted from Steeves et al. (2004), and the greyscale images were manipulated in order to produce the inverted greyscale images. Images were first screened for category member- ship by 25 students, who did not participate in the scene recognition experiment, and only images that were correctly categorized by all stu- dents, in all formats, were included. The size of the images was 256 � 256 pixels, subtending approximately 8.28 � 8.28 of visual angle. There were three categories of natural scenes (coast, forest, and desert) and three categories of non- natural scenes (room, city, and market). Six differ- ent formats were used: normally coloured, inverted colour (each colour was replaced by the opponent colour in colour space, e.g., blue replaced by yellow, green replaced by red), greyscale, inverted greyscale, black-and-white (texture degraded), and spatially rotated images (by 1808 along the horizontal axes, or upside down; see Figure 2 for examples of both a natural and a non-natural scene, used in the experiment, shown in all used formats). Black-and-white images and spatially rotated images were included for full comparison to the findings of Steeves et al. (2004). Here the focus was on the normally coloured images, and these were compared to all other control formats: inverted colour, greyscale, inverted greyscale, black-and-white, and spatially rotated images.

Procedure and design The participants were tested individually in a quiet, darkened room. A trial started with a fix- ation dot for 500 ms, followed by a blank screen for 150 ms, which in turn was followed by an

Table 1. Neuropsychological performance of M.A.H. at 27 months

poststroke

Neuropsychological task Poststroke score

Reasoning

Raven APM 12/12a

WAIS vocabulary 49/60a

Language

Boston Naming Task 180/180a

Token Task 0/21a

Letter Fluency

UNCA 60

Category 39

Memory

RAVLT

Immediate rec. 59/75a

Delayed rec. 14/15a

Recognition 29/30a

Doors Task 21/24a

Rey-O figure delayed rec. 35/36a

WAIS: digit span 19

Location Learning Task

Displacement 14

Learning Index 0.50

Delayed Recall 0

Wechsler Memory Scale: paired

associated

19/21a

Corsi Block Span 6

Executive functioning

Trailmaking

A 33

B 52

Visuoperception & -construction

Judgement of line orientation 28/30a

Face Recognition Task 45

Rey-O figure copy 36/36a

Ishihara test for colour blindness 25/25a

Odd-one-out b

Shape 1.27 c

Luminance 0.46d

Colour 2.16

Motion 2.89 e

Farnsworth_Munsell 100 hue 44 errorsf

Note: WAIS: Wechsler Adult Intelligence Scale; RAVLT: Rey

Auditory Verbal Learning Task; Rey-O: Rey-Osterrieth

Complex Figure. aThe first number is M.A.H.’s score on the test; the second

number is the maximum score. b Data of control participants

on the odd-one-out paradigm: shape, 1.03 mm (SD: 0.68);

luminance, 1.10 cd/m2 (SD: 0.30); colour: 1.28 (SD: 0.78);

motion: 2.33% coherent motion (SD: 1.63).c In mm. dIn cd/

m 2 .

e Percentage of coherent motion.

f Normal range: 30 – 100

errors.

156 COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (2)

NIJBOER ET AL.

image. Each trial was ended via a voice-key switch when the participant gave a verbal response. After a 150-ms pause, the next trial started. Before the experiment started, participants were given the category names of the different scenes along with an example image. They were instructed to name the category to which the image belonged, as quickly and accurately as possible. Verbal responses were recorded with a tape-recorder. No feedback was given during the experiment.

There were two within-subject manipulations: scene type (natural and non-natural) and format (normally colour, inverted colour, greyscale, inverted greyscale, black-and-white, and spatially rotated). Each participant was presented with 360 images (10 images per subscene, all in the six differ- ent formats: 10 images � 6 subscenes � 6 formats).

Data analysis The participants’ responses and reaction times were recorded. All reaction times associated with incorrect responses (less than 2% per participant) were excluded from further analyses. From the overall reaction times for correct responses, trials with reaction times less than 150 ms were excluded, under the assumption that they were anticipatory responses, and reaction times over 2,500 ms were excluded, as they were interpreted as failures of the voice-key measurements. Any reaction times that were more than 2 standard deviations from the participant’s mean were also excluded, as they

were interpreted as outliers. Fewer than 1% of the responses in total were excluded.

The reaction time data of the control parti- cipants were subjected to a two-way repeated measures analysis of variance. When Mauchly’s Test of Sphericity was violated, the Greenhouse – Geisser correction was applied. For planned com- parisons, paired samples t tests with Bonferroni correction for multiple comparisons were used for comparing normally coloured images to all other control formats (inverted colour, greyscale, inverted greyscale, black-and-white, spatially rotated).

Results and discussion

Figure 3 shows the mean reaction times of the control participants. We collapsed the data of the two control groups, resulting in one control group that is very comparable to the one used by Steeves et al. (2004). Moreover, it simplified the methodology to one experiment with one reliable control group. There was no difference in patterns of performance between the two groups of control participants (students and age-matched controls).

As can be seen in Figure 3, there was a signifi- cant main effect of scene type, with shorter overall reaction times for the non-natural scenes than for the natural scenes, F(1, 15) ¼ 6.808, p , .020, which is in line with the data of Steeves et al. (2004). Moreover, there was a significant main effect of format, with the format of the scene

Figure 2. Examples of the stimuli used in the scene recognition experiment (left panel: natural scene; right panel: non-natural scene).

COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (2) 157

EFFECT OF COLOUR AGNOSIA ON SCENE RECOGNITION

image affecting reaction times across scene types, F(5, 75) ¼ 29.827, p , .001. Furthermore, there was a significant interaction between scene type and format, F(5, 75) ¼ 24.584, p , .001: Colour inversion had a large effect on the reaction times of natural scenes, whereas it did not influence non-natural scenes. Reaction times were much longer on the colour inverted images than on nor- mally coloured images for the natural scenes, t(15) ¼ 210.545, p , .001, whereas reaction times on the colour inverted images did not differ from the normally coloured images for the non-natural scenes, t(15) ¼ 21.795, p . .05. Appropriate colour facilitated recognition of natural scenes, as reaction times were shorter for normally coloured images than for all other control formats: black- and-white, t(15) ¼ 29.2164, p , .001; greyscale, t(15) ¼ 25.293, p , .001; inverted greyscale images, t(15) ¼ 28.790, p , .001; and spatially rotated images, t(15) ¼ 27.658, p , .001. For non-natural scenes, reaction times were shorter for normally coloured images only when compared to black-and-white, t(15) ¼ 25.222, p , .001; greyscale, t(15) ¼ 26.241, p , .001; and inverted greyscale images, t(15) ¼ 25.196, p , .010. This effect is also in accordance with Steeves et al. (2004). Importantly, for both natural and non- natural scenes, there was no difference in reaction times between greyscale and inverted greyscale images.

In short, colour manipulations had a large influ- ence on scene recognition for natural images but

not for non-natural images. This is in line with the colour diagnosticity hypothesis. Luminance inversion does not play a role in scene recognition.

Scene recognition in colour agnosia

Having established that it is appropriate colour, not appropriate luminance, that is crucial for rapid scene recognition, we examined M.A.H., a man with developmental colour agnosia. Less than 1% of the data of M.A.H. was excluded due to incorrect responses.

The mean reaction times in the six formats per scene of M.A.H. are presented in Figure 4. As can be seen, there was a significant main effect of scene type, with overall shorter reaction times for the non-natural scenes than for the natural scenes, F(1, 299) ¼ 18.72, p , .001. There was no signi- ficant main effect of format, indicating that the format of the scene image did not affect reaction times irrespective of scene type. In addition, there was no significant interaction.

In order to directly compare the pattern of per- formance of a single case (M.A.H.) on the norm- ally coloured images versus the other control formats with that of the control participants, we used Crawford, Howell, and Garthwaite’s (1998) modified paired samples t test. With this test, we compared the magnitude of the differences between reaction times for colour images and the other formats for M.A.H. with the magnitude of differences between reaction times for colour images and all other formats for the control participants. The planned comparisons using Crawford et al.’s statistics were done with a Bonferroni correction for multiple comparisons.

Overall, M.A.H.’s mean reaction times were longer than those of the control group, for both the non-natural, t(15) ¼ 3.395, p , .010, and the natural scenes, t(15) ¼ 7.251, p , .001.

In spite of his longer reaction times, M.A.H.’s pattern of performance on the non-natural scenes resembled the pattern of performance observed in student and age-matched control participants (see also Figure 3); the magnitude of the differ- ences between normally coloured and all other control formats of M.A.H. fell within the

Figure 3. Mean verbal reaction times (ms) and standard errors of mean (derived from the within-subjects mean square error term) of

the control participants for natural and non-natural scenes in six

different formats.

158 COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (2)

NIJBOER ET AL.

normal range of the control participants (for all comparisons between normally coloured and the other formats used, p . .050). In other words, the effect of colour manipulations on the recog- nition of non-natural scenes is not statistically different between M.A.H. and controls.

M.A.H.’s pattern of performance on the natural scenes, however, differed from that of the control participants. As is apparent from Figure 3, the reaction times of control participants to the natural scenes presented in inverted colour were much longer than those of normally coloured images. For M.A.H., there was no difference in reaction times for images presented in inverted colour than for normally coloured images. This lack of an effect of colour manipulations for M.A.H. is very different from the effect that colour manipulations have on reaction times for the control participants, t(15) ¼ 7.019, p , .001. In addition the absence of a beneficial effect of colour over greyscale and inverted greyscale for M.A.H. was different from the effect of the control participants, t(15) = 3.691, p , .050, and t(15) ¼ 3.269, p , .050, respectively. In other words, the absence of differences between colour and inverted colour, greyscale, inverted greyscale, black-and-white, and spatially rotated images of M.A.H. did not fall within the normal range of the control participants.

Note that the lack of an effect of colour for M.A.H. on the natural scenes condition is not merely an overall inability to name scenes fast and accurately irrespective of format—for example, a floor effect. This notion, tempting though it is,

cannot explain his normal performance on the non-natural scenes. This suggests that M.A.H. does not use colour information to assist him in scene recognition irrespective of colour diagnosticity.

GENERAL DISCUSSION

Colour is an important visual cue. It helps us to decide for instance whether fruit is ripe, and, in a more general sense, it helps us to recognize our environment. Several studies have demonstrated that especially natural scenes and objects are recog- nized better when presented in a colour that is deemed diagnostic for these scenes and objects. Other features, such as contour and visual texture, are especially informative for recognizing non-natural scenes and objects (see below).

The purpose of the present experiments was to examine the role of colour information for scene recognition in colour agnosia. First, we established that the deterioration in scene recognition because of colour inversion found in other studies could not be attributed to the luminance inversion intrinsic to this manipulation. Second, we investigated the level of visual analysis at which colour influences scene recognition by testing a colour agnosic patient. Colour manipulations did not influence the recognition of natural scenes for M.A.H. in the same way that it did for the control participants. Interestingly, M.A.H.’s performance on the natural scenes compared to non-natural scenes was degraded independent of the scene format, as normally coloured and colour-inverted, as well as achromatic, images were responded to with longer reaction times. This could not be attributed to additional processing deficits, such as form or texture (see also Table 1), or an overall inability to name scenes fast and appropriately, as his per- formance on the non-natural scenes was within the range of the control participants.

Which cues are informative for scene recognition?

Obviously, colour is but one of the features for recognizing objects and scenes. Computational

Figure 4. Mean verbal reaction times (ms) of M.A.H. for natural

and non-natural scenes in six different formats.

COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (2) 159

EFFECT OF COLOUR AGNOSIA ON SCENE RECOGNITION

models that permit accurate scene classification without the prior recognition of individual objects (Oliva & Torralba, 2001; Vailaya, Jain, & Zhang, 1998) make use of properties such as the spatial complexity of the scene, visual texture, and colour (Oliva & Schyns, 2000; Oliva & Torralba, 2001). Natural scenes are composed of undulating contours and complex textures, but have characteristic colours. For non-natural scenes, texture and contour cues are more informa- tive, whereas colour cues are not so straight- forward. In other words, colour is the key property for recognizing natural scenes, whereas shape and texture are the key properties for recog- nizing non-natural scenes (Tanaka & Presnell, 1999). Since colour can only become diagnostic when one learns that certain objects or scenes tend to have certain colours, M.A.H., who has developmental colour agnosia, never learned this diagnostic value and thus lacks the benefit of colour information for recognizing natural scenes. His poorer internal representation of colour (e.g., impaired colour imagery, conceptualiz- ation, and explicit associations; Nijboer et al., 2006; van Zandvoort et al., in press), due to his develop- mental colour agnosia together with only texture and contour cues available for recognizing the scenes, resulted in his overall slower, but still very accurate, recognition of natural scenes. In other words, it is the specific combination of an impaired association between natural scenes and specific colours and the spatial complexity (complex textures and contours) of these natural scenes that can explain his overall slowing in recognizing the images.

Based on M.A.H.’s performance on the scene recognition task, one would expect a patient with visual form agnosia to show the inverse pattern—that is, slower performance on the non- natural scenes. Interestingly, this is exactly what has been found in the study of Steeves et al. (2004; see their Figure 4), where D.F. was much slower in identifying non-natural scenes than natural scenes. These findings are in line with the results of Tanaka and Presnell (1999) who found that when shape information is degraded, recognition of objects low in colour diagnosticity

is more impaired than recognition of objects high in colour diagnosticity.

Levels of visual processing

An important reason for testing scene recognition in colour agnosia is that there still is considerable debate about the level of visual analysis at which colour influences recognition of objects and scenes. By testing a colour agnosic patient, we can distinguish between these levels of analysis. Colour associations and semantic representations are selectively impaired or lost, but low-level colour perception is not affected in colour agnosia (van Zandvoort, Nijboer, & de Haan, in press). Any deviation in performance on the scene recognition task is therefore more likely to be attributable to a deficit in processing of colour information at a higher cognitive level than to a deficit at a sensory level. Wurm et al.’s (1993) con- clusion of a sensory role for colour was based upon results from experiments in which all objects would, according to the definition used in our as well as Price and Humphrey’s (1989) experiments, make up a colour diagnostic stimulus set. Apparently, the level of diagnosticity (high or low, e.g., carrot or apple, respectively) does not influence the beneficial effect of presenting objects in their diagnostic colour. Note, however, that our conclusion about M.A.H.’s performance does not depend upon either the definition of colour diagnosticity or the stimulus set used (see Introduction).

Therefore we can conclude, in line with Price and Humphreys (1989), that the characteristic colour of an object or scene is stored in the structural or semantic representation of that object or scene. Apparently, the surface property colour linked to specific (i.e., natural) objects or scenes is stored in memory. The beneficial effect of appropriate colour in recognizing natural scenes occurs at a higher cog- nitive level and is the result of internal colour represen- tations, based on learned associations.

Manuscript received 16 February 2006

Revised manuscript received 3 August 2006

Revised manuscript accepted 31 August 2006

First published online 5 December 2006

160 COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (2)

NIJBOER ET AL.

REFERENCES

Biederman, I., & Ju, G. (1988). Surface versus edge- based determinants of visual recognition. Cognitive Psychology, 20, 38 – 64.

Crawford, J. R., Howell, D. C., & Garthwaite, P. H. (1998). Payne and Jones revisited: Estimating the abnormality of test score differences using a modified paired samples t test. Journal of Clinical and Experimental Neuropsychology, 20, 898 – 905.

De Haan, E. H., Heywood, C. A., Young, A. W., Edelstyn, N., & Newcombe, F. (1995). Ettlinger revisited: The relation between agnosia and sensory impairment. Journal of Neurology, Neurosurgery, and Psychiatry, 58, 350 – 356.

De Renzi, E., & Faglioni, P. (1978). Normative data and screening power of a shortened version of the Token Test. Cortex, 14, 41 – 49.

Farnsworth, D. (1957). The Farnsworth_Munsell 100-Hue test for the examination of color vision. Baltimore, MD: Munsell Color Company.

Humphrey, G. K., Goodale, M. A., Jakobson, L. S., & Servos, P. (1994). The role of surface information in object recognition: Studies of a visual form agnosic and normal subjects. Perception, 23, 1457 – 1481.

Ishihara, S. (1977). Tests for colour-blindness, 38 plates edition. Tokyo: Kanehara Shuppan Co.

Joseph, J. E., & Proffitt, D. R. (1996). Semantic versus perceptual influences of color in object recognition. Journal of Experimental Psychology. Learning,

Memory, and Cognition, 22, 407 – 429. Livingstone, M. S., & Hubel, D. H. (1987).

Psychophysical evidence for separate channels for the perception of form, color, movement, and depth. Journal of Neuroscience, 7, 3416 – 3468.

Mapelli, D., & Behrmann, M. (1997). The role of color in object recognition: Evidence from visual agnosia. Neurocase, 3, 237 – 247.

Milner, A. D., & Heywood, C. A. (1989). A disorder of lightness discrimination in a case of visual form agnosia. Cortex, 25, 489 – 494.

Milner, A. D., Perrett, D. I., Johnston, R. S., Benson, P. J., Jordan, T. R., Heeley, D. W., et al. (1991). Perception and action in “visual form agnosia”. Brain, 114, 405 – 428.

Nijboer, T. C. W., van Zandvoort, M. J. E., & de Haan, E. H. F. (2006). Covert colour processing in colour agnosia. Neuropsychologia, 44, 1437 – 1443.

Oliva, A., & Schyns, P. G. (1997). Coarse blobs or fine edges? Evidence that information diagnosticity changes the perception of complex visual stimuli. Cognitive Psychology, 34, 72 – 107.

Oliva, A., & Schyns, P. G. (2000). Diagnostic colors mediate scene recognition. Cognitive Psychology, 41, 176 – 210.

Oliva, A., & Torralba, A. (2001). Modeling the shape of the scene: A holistic representation of the spatial envelope. International Journal of Computer Vision, 42, 145 – 175.

Ostergaard, A. L., & Davidoff, J. B. (1985). Some effects of color on naming and recognition of objects. Journal of Experimental Psychology. Learning, Memory, and Cognition, 11, 579 – 587.

Price, C. J., & Humphreys, G. W. (1989). The effects of surface detail on object categorization and naming. Quarterly Journal of Experimental Psychology, 41, 797 – 827.

Steeves, J. K. E., Humphrey, G. K., Culham, J. C., Menon, R. S., Milner, A. D., & Goodale, M. A. (2004). Behavioral and neuroimaging evidence for a contribution of color and texture information to scene classification in a patient with visual form agnosia. Journal of Cognitive Neuroscience, 16, 1 – 11.

Tanaka, J. W., & Presnell, L. M. (1999). Color diagnos- ticity in object recognition. Perception & Psychophysics, 61, 1140 – 1153.

Vailaya, A., Jain, A., & Zhang, H. J. (1998). On image classification: City images vs. landscapes. Pattern Recognition, 31, 1921 – 1935.

van Zandvoort, M. J. E., Nijboer, T. C. W., & de Haan, E. H. F. (in press). Developmental colour agnosia. Cortex.

Wurm, L. H., Legge, G. E., Isenberg, L. M., & Luebker, A. (1993). Color improves object recog- nition in normal and low vision. Journal of Experimental Psychology. Human Perception &

Performance, 19, 899 – 911.

COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (2) 161

EFFECT OF COLOUR AGNOSIA ON SCENE RECOGNITION