Colour and the Processing of Chromatic Information
Abstract and Keywords
This chapter discusses colour discrimination and other tasks involving the visual the processing of spectral information. It focuses on chromatic habituation, a technique used to isolate colour detection mechanisms. Colour detection mechanisms have sensitivities that lie between those of standard black standard black-white, red-green, and yellow-blue mechanisms. A study of the macaque visual cortex shows that with chromatic habituation, those mechanisms with intermediate sensitivities determine behaviour in practical visual search and colour detection tasks.
My interest in colour vision was formed during my graduate work with Peter Lennie at the University of Rochester, New York, during the mid-1980s. Peter Lennie and John Krauskopf were studying chromatic processing in macaque lateral geniculate nucleus (LGN) using electrophysiological techniques, and I fell naturally into related research using psychophysical techniques.
Krauskopf ’s psychophysical experiments with chromatic habituation in the early 1980s revealed cardinal axes for colour vision which matched peak chromatic sensitivities in the LGN. The cardinal axes do not match the colour-opponent mechanism sensitivities found with judgements of colour appearance, like those of Hurvich and Jameson (1955), and the question arose of the relationship between detection-based cardinal axis sensitivities and appearance-based colour-opponent mechanism sensitivities. The thinking is that experiments involving detection are more objective than those involving judgements of appearance, and the underlying question was thus whether one could study the mechanisms involved in the conscious representation of colour using objective techniques.
Work at that time with noise-masking techniques to study achromatic vision (particularly that of Denis Pelli) prompted me to study this question using noise masking. I thought that perhaps the standard red–green and yellow–blue mechanism sensitivities could be measured in a detection experiment using a noise-masking variant of Stiles’ techniques. For instance, a red signal would be visible to only the red–green mechanism, and threshold elevation by noise along various axes in colour space could be used to measure its sensitivity. Experimental results seemed to confirm the hypothesis that, in detection tasks, signals of unique hue could be used to reveal the sensitivities of appearance-based mechanisms. Yet the results found with signals of intermediate hue, such as orange, suggested that observers detect such a signal by using a mixture of mechanisms that depends on the chromatic properties of the noise mask. For instance, one can detect an orange signal presented in yellow–blue noise using a ‘red’-sensitive mechanism, but one detects the same signal using a ‘yellow’-sensitive mechanism when it is presented in red–green noise.
This ‘off-axis looking’ led me to formulate a model in which noise is represented by a covariance matrix, the inversion of which reveals the chromatic sensitivity of the mechanism with the greatest signal-to-noise ratio. Such a model, I felt, would quantify observers’ tendency to detect a signal of intermediate hue using a mechanism sensitive to the signal yet minimally sensitive to a particular noise. Donald MacLeod pointed out to me that such a model is isotropic, and cannot distinguish between unique-hue stimuli and those of intermediate hue. The model suggests that observers should use varying detection mechanisms to detect not only an orange signal but also unique red, unique green, unique yellow, and unique blue signals. Yet the results of my experiments did not reveal off-axis looking for unique-hue signals.
I went back to the drawing board. I suspected that observers may have been able to improve their sensitivity to unique-hue signals using off-axis looking, but was not willing to redo the noise-masking experiments, as they are quite tedious and involve looking for things which typically cannot be seen. I turned to the visual search paradigm used by Anne Treisman, which was in vogue among vision researchers at the time. The results of these experiments were consistent with the isotropic model; they did not reveal the standard colour-opponent mechanism sensitivities. These and further experiments are discussed in this chapter.
(p. 144 ) Introduction
The study of colour vision has focused on the relation between colour appearance and the physical stimulation of the eye by light. Newton (1704/1952) spread sunlight into a visible spectrum using a prism and so linked light wavelength and perceived hue. Young (1802) and Helmholtz (1909/1962) accounted for colour-matching data by developing the trichromatic theory of colour vision, which held that perceived colour is determined by the extent to which a light stimulates three classes of retinal photoreceptor. The work of Hering (1878/1964) and, later, Hurvich and Jameson (1957), completed the present, standard account of colour vision by adding a second, colour-opponent, stage of chromatic processing that transforms signals from the initial photoreceptoral stage. In this view, the appearance of a light is determined by the extent to which it stimulates three classes of colour-opponent neurons, rather than the extent to which it stimulates three classes of photoreceptors.
Trichromatic and colour-opponent theories have been used to account for how our visual system uses spectral information. One such use is to discriminate between two lights on the basis of a difference in their spectral properties. Helmholtz (1891, 1892) used a line-element model which measures the difference in stimulation of the three classes of photoreceptor to fit wavelength discrimination data. The activities of the photoreceptors are then scaled to quantify the two lights’ discriminability (von Kries 1905). Other trichromatic line-element models scale and combine photoreceptoral information in different ways (e.g. Stiles 1946).
Colour-opponent theory has also resulted in its share of line-element and related models of colour discrimination (e.g. Hurvich and Jameson 1955; Vos and Walraven 1972a, b). The aim, again, has been to fit colour discrimination data by supposing that activity determining colour appearance also determines discriminability. This work shows that outputs from colour-opponent units must be scaled to take into account adaptation, viz. the regulation of visual sensitivity. Indeed, there was one point in the history of colour science when there was strong evidence for unitary mechanisms involved in colour discrimination, Stiles’ ∏ mechanisms, that corresponded neither to the three classes of photoreceptor nor to the black–white, red–green, and yellow–blue colour-opponent processes (Stiles 1978). Yet these mechanisms were soon understood in terms of the adaptation of the standard colour-opponent mechanisms (Pugh 1976).
In summary, accounts of colour discrimination and other tasks that involve the visual processing of spectral information have been derived historically from accounts of colour appearance. Indeed, all is well in the world of colour research when performance in chromatic processing tasks can be understood as a by-product of colour appearance.
The problem is that psychophysical and electrophysiological research in colour vision no longer supports this simple view. This chapter reviews briefly work showing that we use mechanisms, in colour detection tasks, which have sensitivities that lie between those of the standard black–white, red–green, and yellow–blue mechanisms. Discovered by Krauskopf and his colleagues, working with chromatic habituation (Krauskopf et al. 1986), these mechanisms with intermediate sensitivities have a clear anatomical correlate in the macaque visual cortex and have been shown to determine behaviour in practical visual search and colour detection tasks.
(p. 145 ) Chromatic habituation
Krauskopf and colleagues used a habituation technique to isolate colour detection mechanisms. Observers in their experiments exposed themselves to high-contrast modulations of chromaticity along various directions in colour space. Prolonged exposure to the high-contrast modulations desensitizes observers in a colour-selective way. For instance, if observers stare at a pulsing red–green stimulus for 1 min, their ability to detect red or green stimuli in the immediately following period is poorer, but their ability to detect yellow or blue stimuli is relatively unimpaired. Likewise, staring at a yellow–blue pulsation impairs the detection of yellow or blue signals but has little effect on the detection of red or green signals. The interpretation of these results is that observers possess both red–green and yellow–blue opponent mechanisms, and that these can be desensitized or habituated independently of one another (Krauskopf et al. 1982).
The interesting part arose when Krauskopf and colleagues asked their observers to view pulsing stimuli with intermediate colours, such as orange. If observers have only the red–green and yellow–blue colour-opponent mechanisms, then one would expect that a pulsing orange–blue-green stimulus will desensitize both the red–green and the yellow–blue mechanisms: detection of all colours will be equally impaired. In fact, the ability of observers to detect chromatic signals was impaired most along the orange–blue-green direction in colour space, and least along a perpendicular, yellow-green–violet direction. This result suggests that observers possess colour-opponent mechanisms with hue sensitivities that lie between those of the standard mechanisms (Krauskopf et al. 1986). These observations have since been extended in a variety of work with the chromatic habituation paradigm (see Webster Chapter 2, this volume).
Monkey visual cortex
Studies of colour-sensitive neurons in the macaque visual cortex paint a picture that is consistent with the activity of mechanisms with intermediate hue sensitivities. At the time when Krauskopf and colleagues performed their studies, little quantitative work had been done on chromatic sensitivity in the macaque cortex. More work had been done with neurons in the lower levels of the monkey visual system, including retina and lateral geniculate nucleus (LGN). Studies of the LGN indicate an organization that is partially consistent with, but not identical to, the standard colour-opponent scheme (DeValois et al. 1966; Derrington et al. 1984). The chromatic sensitivities of parvocellular colour-sensitive neurons in the LGN cluster around two axes in colour space that correspond only roughly to what one would expect of a red–green and a yellow–blue mechanism. One cannot identify these neurons with the standard colour-opponent channels for two reasons. First, the spectral sensitivity of the red–green mechanism suggested by studies of the LGN does not match the sensitivity deduced psychophysically. The putative red–green neurons tend to have no input from short-wavelength-sensitive cones, yet standard accounts of colour-opponency invoke a short-wavelength-sensitive cone contribution to redness, in order to help account for the redness seen within violet lights of short wavelength (Hassenstein 1968; Jameson and Hurvich 1968; D'Zmura 1990). Secondly, many colour-sensitive neurons in the LGN (p. 146 ) are also sensitive to black–white modulation; the red–green and yellow–blue mechanisms of standard colour-opponent theory are insensitive to black–white modulation (Ingling and Martinez-Uriegas 1983).
One may combine signals from neurons in the LGN to create colour sensitivities that are like those required by the standard theory (D'Zmura and Lennie 1986). As shown in Fig. 4.1, the addition of signals from colour-sensitive neurons in the LGN can provide neurons insensitive to achromatic signals but with chromatic sensitivities that match red–green and yellow–blue. Figure 4.1 also shows that adding such signals can give rise to neurons with intermediate spectral sensitivities—neurons that are tuned to orange, yellow-green, blue-green, and violet directions in colour space.
Work over the past decade has shown conclusively that the visual cortex of macaque monkeys is replete with colour-sensitive neurons with hue sensitivities that lie intermediate
Colour in visual search
Visual search experiments provided the first evidence that the mechanisms tuned to intermediate hues are active in everyday colour detection tasks (D'Zmura 1991).
Observers sought an orange disc placed among distractor discs with different colours. The Treisman visual search paradigm (Treisman and Gelade 1980) was used to determine how effectively the distractors hindered the search for the orange target disc. Four experimental conditions that correspond to different colours for the distractor discs are shown diagramatically in Fig. 4.2. The colour of the orange disc is shown in the colour plane by the unfilled circle in each of the panels. Distractor colours are shown by the filled circles.
Figure 4.2a refers to an experimental condition in which the orange disc is presented among red and green distractors. Results showed that the search for an orange target among red and green distractors occurs spatially in parallel across the central visual field. The orange target ‘pops out’ of the field of red and green distractors. The yellow–blue (p. 148 ) opponent mechanism is a plausible candidate for detection under these circumstances. The yellow–blue mechanism can pick up the yellow within the orange target disc, but is completely insensitive to the red and green distractor discs.
The experimental condition in which the orange disc is presented among yellow and blue distractors is shown diagrammatically in Fig. 4.2b. Again, the orange target pops out. The red–green opponent mechanism is a plausible candidate for detection in this condition: it will pick up the red within the orange target disc, but will be completely insensitive to the yellow and blue distractor discs.
What happens if the orange target is displayed among yellow-green and violet distractors, as shown in Fig. 4.2c? The yellow–blue mechanism cannot detect the orange target reliably, because the mechanism is also sensitive to the yellow-green distractors. Likewise, the red–green mechanism cannot detect the orange target reliably, because it is sensitive to the redness within the violet distractors. Nevertheless, the orange target pops out. The simplest explanation is that we possess a detection mechanism tuned to orange that is simultaneously insensitive to yellow-green and violet.
The spectral width of the orange-sensitive mechanism is tested in the condition shown Fig. 4.2d. If the distracting colours are brought too close to the colour of the orange target, then what was formerly an easy, parallel search becomes a difficult search, most likely conducted in a serial fashion. Evidently, the orange mechanism used under the conditions of Fig. 4.2c does not have a narrow sensitivity, because crowding the orange target with red and yellow distractors makes the search difficult.
The pattern of results suggests that the colour mechanisms used in parallel search have broad, linear spectral sensitivities. In the parallel search conditions, observers can set a threshold on a linear mechanism that distinguishes the target from the distractors. These thresholds are indicated in Fig. 4.2A–C by the lines marked ‘T’ that separate the orange target from the various sets of distractors. This pattern of findings was shown to hold true for red, yellow, orange, and violet targets. It provides conclusive evidence for the activity of mechanisms with intermediate hue sensitivities in everyday detection tasks (D'Zmura 1991).
Further work on colour in visual search by Bauer et al. (1996, 1998) and by Olds et al. (1999) has replicated and extended these results in several interesting ways. The work by Olds and colleagues has shown, in particular, that a parallel search for a target of intermediate hue does not depend on an observer's prior knowledge of target and distractor colours. The mechanisms sensitive to intermediate hues are not created through top-down awareness. They are simply there, ready to go.
D'Zmura et al. (1997) showed that these mechanisms can operate at a high level of visual processing, after figure/ground segmentation has taken place. Shaded, coloured discs were used in visual search experiments. The discs were shaded in such a way that some of them, positioned randomly, appeared within the perceived figure, whereas others appeared in the ground. An orange target disc was placed among yellow-green and violet distractors in the figure. Orange distractor discs were placed in the ground. Results showed that the search for the orange target among the orange, yellow-green and violet distractors was conducted spatially in parallel. The orange distractors in the ground provided no problem whatsoever, presumably because the colour-based search using the ‘orange’ mechanism occurred within a segmented representation of the display.
(p. 149 ) Colour detection
The mechanisms sensitive to intermediate hues function not only at the suprathreshold stimulus levels used in the visual search experiments, but also at parathreshold levels common in standard detection experiments. This was shown in a recent noise-masking study that measured the spectral properties of chromatic detection mechanisms (D'Zmura and Knoblauch 1998).
Earlier noise-masking studies used a simple kind of noise called ‘axial’ noise, which is a random modulation of colour along a particular axis in the colour plane. For instance, a light that flickers randomly among various lights along the red–green axis is red–green axial noise. One expects, correctly, that detecting a faint red signal pulse will be made more difficult if one adds a sufficient amount of red–green axial noise to the signal (D'Zmura 1990). Unfortunately, one cannot use axial noise to measure the spectral properties of detection mechanisms. Consider an experiment, similar to that shown in Fig. 4.2, in which a faint orange signal is masked by axial noise. One might think that one can measure the spectral sensitivity of the orange detection mechanism by measuring how well noise, along systematically varied axes, masks the orange signal. This strategy does not work. For example, by masking the orange signal with red–green noise, one forces the observer to use the ‘yellow’ mechanism to detect the orange stimulus. By masking the orange signal with yellow–blue noise, one forces the observer to use the ‘red’ mechanism. Clearly, the mechanism that one uses to detect a single chromatic signal depends on the choice of noise chromatic properties, and measuring the spectral sensitivity of a single detection mechanism in this manner is impossible.
D'Zmura and Knoblauch (1998) used a new kind of noise, ‘sectored’ noise, to overcome this weakness of earlier studies with axial noise. As shown in Fig. 4.3, noise samples are drawn from a sector in colour space that is centred on an axis a of the colour signal to be detected. One varies the sector width and measures how well the noise masks the signal. The technique borrows from earlier work in studies of ‘critical bands’ in audition (Fletcher 1940).
By measuring the potency of noise masking as a function of sector width, one can distinguish three possibilities:
(1) that the signal is detected by a linear detection mechanism that matches the axis of the signal;
(2) that the signal is detected by a mechanism with a narrow spectral sensitivity that matches the axis of the signal; and
(3) that the signal is detected by two standard colour-opponent mechanisms with sensitivities that do not depend on the choice of signal.
In the first case, varying sector width has no effect on signal detection. In the second case, increasing sector width causes noise samples to become increasingly less effective in stimulating the detection mechanism, because the noise samples fall outside the narrow region of sensitivity. In the third case, increasing sector width can cause the noise to grow more effective, because noise samples fall increasingly in areas to which the standard mechanisms are most sensitive. (p. 150 )
Experimental results show that noise masking is independent of sector width, for signals that appear yellow, orange, red, and violet. One infers that observers use linear detection mechanisms tuned to these hues. The ready generalization is that colour detection is served by mechanisms tuned to a variety of directions in colour space, and that these mechanisms have broad, linear spectral sensitivities (D'Zmura and Knoblauch 1998).
Colour appearance is available only through conscious representation. Yet most would agree that consciousness does not provide a complete picture of all neural processing. For instance, consider a machine vision system that uses chromatic information to perform a task such as colour quality control on a factory assembly line. One would be hard pressed to claim that such a system performs its task using colour representations in consciousness. There is no reason to assume that colour appearance, available only through consciousness, can be used to understand performance in chromatic processing tasks.
(p. 151 ) This important point has been made repeatedly by Hurvich, who has reminded many speakers in public meetings that chromatic processing must be distinguished from colour appearance. Yet Hurvich and Jameson have had difficulty in articulating this position, because the mechanisms that serve chromatic processing and those that serve colour appearance are isomorphic in their colour-opponent theory (Hurvich and Jameson 1957).
The experimental work that has been reviewed in this chapter makes clear that detection mechanisms are not structured according to standard colour-opponent theory. Rather, the peak spectral sensitivities of detection mechanisms in habituation, visual search, and noise-masking tasks are scattered uniformly in the colour plane. We can ask, conversely: do the mechanisms that are directly responsible for colour appearance have this multiple-mechanism sort of organization rather than the standard red–green and yellow–blue organization? There is good reason to think not. The unique hues provide robust psychological evidence for the standard colour-opponent organization. Unique red, yellow, green, and blue have a unitary, fundamental quality, while orange, yellow-green, blue-green, and violet are composite in nature. Although the multiple-mechanism organization can be forced to provide unique hues, this can only be accomplished in a post hoc fashion.
The evidence suggests, then, that we must distinguish carefully between colour and the processing of chromatic information. Different organizations of colour sensitive mechanisms appear to underlie behaviour concerning colour appearance and behaviour in detection tasks.
Bauer, B., Jolicoeur, P., and Cowan, W. B. (1996). Visual search for colour targets that are or are not linearly separable from distractors. Vision Research 36, 1439–1465.
Bauer, B., Jolicoeur, P., and Cowan, W. B. (1998). The linear separability effect in colour visual search: Ruling out the additive-colour hypothesis. Perception and Psychophysics 60, 1083–1093.
Derrington, A. M., Krauskopf, J., and Lennie, P. (1984). Chromatic mechanisms in lateral geniculate nucleus of macaque. Journal of Physiology (London) 357, 241–265.
DeValois, R. L., Abramov, I., and Jacobs, G. H. (1966). Analysis of response patterns of LGN cells. Journal of the Optical Society of America 56, 966–977.
DeValois, R. L., Cottaris, N., and Elfar, S. (1997). S-cone inputs to striate cortex cells. Investigative Ophthalmology and Visual Science 38, 15.
D'Zmura, M. (1990). Surface color psychophysics. PhD Dissertation, University of Rochester, New York.
D'Zmura, M. (1991). Color in visual search. Vision Research 31, 951–966.
D'Zmura, M. and Knoblauch, K. (1998). Spectral bandwidths for the detection of color. Vision Research 38, 3117–3128.
D'Zmura, M. and Lennie, P. (1986). Mechanisms of color constancy. Journal of the Optical Society of America A 3, 1662.
D'Zmura, M., Lennie, P., and Tiana, C. (1997). Color search and visual field segregation. Perception and Psychophysics 59, 381–388. (p. 152 )
Fletcher, H. (1940). Auditory patterns. Reviews of Modern Physics 12, 47.
Gegenfurtner, K. R., Kiper, D. C., and Levitt, J. B. (1997). Functional properties of neurons in macaque area V3. Journal of Neurophysiology 77, 1906–1923.
Hassenstein, B. (1968). Modellrechnung zur Datenverarbeitung beim Farbensehen des Menschen. Kybernetik 4, 209–223.
Helmholtz, H. von (1909/1962). Treatise on physiological optics, (3rd edn), (ed. J. P. C. Southall). Dover, New York.
Helmholtz, H. von (1891). Versuch einer erweiterten Anwendung des Fechnerschen Gesetzes im Farbensystem. Zeitschrift für Psychologie und Physiologie der Sinnesorgane 2, 1.
Helmholtz, H. von (1892). Versuch, das psychophysische Gesetz auf die Farbenunterschiede trichromatischer Augen anzuwenden. Zeitschrift für Psychologie und Physiologie der Sinnesorgane 3, 1.
Hering, E. (1878/1964). Outlines of a theory of the light sense (translated by L. M. Hurvich and D. Jameson). Harvard University Press, Cambridge, MA.
Hurvich, L. M. and Jameson, D. (1955). Some quantitative aspects of an opponent-colors theory. II. Brightness, saturation, and hue in normal and dichromatic vision. Journal of the Optical Society of America 45, 602–616.
Hurvich, L. M. and Jameson, D. (1957). An opponent-process theory of color vision. Psychological Review 64, 384–404.
Ingling, C. R., Jr, and Martinez-Uriegas, E. (1983). The relationship between spectral sensitivity and spatial sensitivity for the primary, r, g x-channel. Vision Research 23, 1495–1500.
Jameson, D. and Hurvich, L. M. (1968). Opponent response functions related to measured cone photopigments. Journal of the Optical Society of America 58, 429–430.
Kiper, D. C., Fenstemaker, S. B., and Gegenfurtner, K. R. (1997). Chromatic properties of neurons in macaque area V2. Visual Neuroscience 14, 1061–1072.
Krauskopf, J., Williams, D. R., and Heeley, D. W. (1982). Cardinal directions of color space. Vision Research 22, 1123–1131.
Krauskopf, J., Williams, D. R., Mandler, M., and Brown, A. (1986). Higher-order color mechanisms. Vision Research 26, 23.
Kries, J. von (1904). Die Gesichtsempfindungen. In Handbuch der Physiologie des Menschen, Vol. 3, (ed. W. Nagel), pp. 109–279. Vieweg, Braunschweig.
Lennie, P., Krauskopf, J., and Sclar, G. (1990). Chromatic mechanisms in striate cortex of macaque. Journal of Neuroscience 10, 649–669.
Newton, I. (1704/1952). Optics. In Great books of the Western world, Vol. 34 (ed. R. M. Hutchins), pp. 377–544. William Benton, Chicago.
Olds, E. S., Cowan, W. B., and Jolicoeur, P. (1999). Stimulus-determined mechanisms for color search. Perception and Psychophysics 61, 1038–1045.
Pugh, E. N., Jr (1976). The nature of the ∏1 colour mechanism of W. S. Stiles. Journal of Physiology (London), 257, 713–747.
Stiles, W. S. (1946). A modified Helmholtz line-element in brightness-colour space. Proceedings of the Physical Society 58, 113–137.
Stiles, W. S. (1978). Mechanisms of colour vision. Academic Press, New York.
Treisman, A. M. and Gelade, G. (1980). A feature-integration theory of attention. Cognitive Psychology 12, 97–136.
Vos, J. J. and Walraven, P. L. (1972a). An analytical description of the line element in the zone-fluctuation model of colour vision—I. Basic concepts. Vision Research 12, 1327.
Vos, J. J. and Walraven, P. L. (1972b). An analytical description of the line element in the zone-fluctuation model of colour vision—II. The derivation of the line element. Vision Research 12, 1345.
Young, T. (1802). On the theory of light and colours. Philosophical Transactions of the Royal Society London, 92, 12–48.
Commentary on D'Zmura
In his chapter, D'Zmura presents a lucid and intriguing review of the connections between theories of chromatic discrimination and theories of colour appearance. The best preparation before reading it would be to recall that psychophysics originated in Gustav Fechner's attempt to explain the apparent intensity of sensory variables, such as lightness, in terms of the observer's ability to discriminate intensities. This derivation of Fechner's law from Weber's has been challenged repeatedly, in particular by S. S. Stevens (Stevens 1957; Roberts 1979, pp. 149–184) and this controversy alone should alert us to the possibility that there is no simple relation between human chromatic discrimination performance and human judgements of colour appearance.
D'Zmura’s point is important. To concentrate on colour appearance alone, as so many of the other chapters in this volume do, is analogous to treating statistics as a collection of methods for estimation, ignoring the statistical machinery that is appropriate and relevant for hypothesis testing. One can push this analogy a bit further.
In the context of colour vision, ‘estimation’ amounts to assigning colours to the pieces that make up the world. It is evidently difficult to estimate colour so that assigned colours correspond to surface properties, the problem of colour constancy. But the goal of the estimation problem is clear. Hypothesis testing is manifold: it can be used to decide whether a specific piece of the world has changed recently, whether one piece is different from another, whether something seen a while ago has returned, or whether one piece really doesn't fit in very well with the rest. There are psychophysical procedures used in colour vision corresponding to each sort of hypothesis test just outlined. The tests involve comparison across space and time, detection of visual transients, and detection of anomalies in complex arrays of stimulus elements. It is evidently a challenge to develop a computational model of colour processing that combines estimation and all the different kinds of neural ‘machinery’ needed for testing the kinds of hypotheses just outlined. D'Zmura emphasizes that there is psychophysical and neurophysiological evidence implying that the representations underlying colour appearance and chromatic discrimination are remarkably different. He describes his work in collaboration with Knoblauch, involving an elegant noise-masking method (analogous to that used in critical-band experiments in audition), as well as his work in the visual search for coloured targets. It is impressive that the conclusions drawn from two psychophysical tasks that are, in appearance, unrelated are so similar, and that this common conclusion is in agreement with the neurophysiological evidence: the distribution of mechanisms encoding colour information is roughly uniform in colour space. There are no distinguished directions in colour space corresponding to those isolated by experiments concerning colour appearance. D'Zmura makes his point very well.
It's not appropriate, though, to talk about a representation without also specifying the neural machinery that links it to psychophysical judgements of various kinds. As D'Zmura demonstrates, simple weighted linear sums can be used to create virtual mechanisms that can have any direction in colour space. What sort of role do spontaneously created virtual mechanisms of this sort have in carrying out the psychophysical tasks he describes, that correspond to different sorts of hypothesis tests? Indeed, what exactly happens when we compare two locations separated in space or in time, or try to detect the ‘odd man out’ in a visual search experiment? It is possible that estimation and hypothesis testing tasks can share a common representation and yet differ in the neural operations that each task employs. For example, the visual system may represent colour information interchangeably in many directions in colour space, but a critical piece of neural machinery required to carry out hue cancellation may not operate equally well along arbitrary directions in colour space.
D'Zmura’s chapter is a succinct and well-written summary of his own work and the work of others concerning the neural representation of colour and what we can earn about it by psychophysical means.