Here are abstracts for the October 2002 phase of the conference Color Perception: Philosophical and Scientific Perspectives.
Color Blindness, Color Language and the Dimensions of Color Appearance
Dichromats apparently possess two types of cone in the retina instead of the normal three, and, on a standard view, their color vision can therefore have only two dimensions rather than the normal three. The space representing hue, saturation and brightness collapses to a single plane: the dichromat supposedly sees no more than two hues, with variations in their saturation and lightness -- e.g. (as in Judd 1948) a yellow of ~575 nm and a blue of ~470 nm. What then is going on when dichromats use terms like 'red', 'green', 'orange' and 'purple', if they can really only see one yellow hue and one blue? And what is going on when (as in Jameson & Hurvich 1978) some dichromats even apply such terms correctly to the colored caps in the Farnsworth D-15 test?
There are many factors to consider (lightness cues, rod inputs, seeing a surface from various angles as it reflects light of various spectral profiles), as well as studies of cases of unilateral color vision deficits. I shall say something about the projection of 2-dimensional color-data into 3-dimensional space, and about aspect-shift in color-perception -- a special case of the 'interpretation' of ambiguous stimuli. A survey of a variety of philosophical views of the nature of colors (and in particular, surface colors) suggests that we may even be able to allow the dichromat to have a proper conception of the colors whose spectral equivalents he cannot in the normal way see. All of which suggests some new directions for experimental investigation, which, as a philosopher, I am sorry not to be qualified to perform.
Some Philosophical Applications of Variation in Color Vision
Philosophers have often taken variation in color vision to support significant claims about color and/or perception. Sometimes this variation is merely possible, as in John Locke's "inverted spectrum"; sometimes it is actual variation, as in C. L. Hardin's use of Hurvich and Jameson on the varying loci of unique-green. The significant claims include the view that nothing is colored, and the view that colors are relative to perceivers. I shall briefly explain and discuss the arguments for some of these claims.
Conceptually interesting variations in color vision can be produced in normal observers by subjecting them to visual search tasks that are quite demanding, or by masking stimuli almost immediately after they are presented. These are two experimental routes into the study of preattentive vision. There is some evidence that the preattentive representation of color differs dramatically from that found in normal color vision. In particular, the masking studies are conceptually interesting because they allow us to drive a wedge quite decisively between "mental states with phenomenal character" and "conscious mental states." I will review one such study and show how it provides evidence for the existence of unconscious color sensations; that is, for chromatic sensory states that have phenomenal character yet which are in every way unconscious.
John S. Werner
Representation of Whiteness with Changing Retinal Stimulation
The human lens is clear in the newborn eye and gradually decreases in its light transmission, particularly at short wavelengths. As a result, the illumination of the retina gradually changes across the life span. Concurrently, there is a reduction in sensitivity of the photoreceptors with increasing age, albeit with substantial individual differences in the rate of change. As a result, there is a great deal of individual and age-related variation in the retinal stimulus and the signals leaving the retina for perceptual analysis. Nevertheless, we have found that there is a remarkable degree of stability in color perception across the life span. We have previously shown that the white point (and the wavelengths of unique blue and yellow) is relatively invariant with normal aging, implying that the visual system renormalizes itself to maintain constancy of perception. Thus, while we may call the same stimulus "white" as we did 40 years ago, it is based upon rather different retinal stimulation. With colleagues Peter Delahunt and Mike Webster, we observed a shift in the white point following cataract surgery (removal of a brunescent lens) that was initially toward yellow in color space, but over the course of months, it returned nearly to what it was in the presence of a cataractous lens. This long-term adaptation is as it should be; otherwise, the white of the young would be the blue of the old. Renormalization to white, rather than other colors, may be particularly useful as an internal reference for color constancy.
Kimberly A. Jameson
Color Perception in Observers Multiple Photopigment Opsin Gene Polymorphisms
Unexpected differences have been found in the color perception of observers possessing the genetic potential for four or more classes of retinal photopigments (Jameon, Highnote & Wasserman, 2001. Psychonomic Bulletin & Review). Using novel multidimensional scaling (MDS) methods, we compared color vision assessment results of individuals possessing tetrachromatic retinal genotypes with performance from individuals with trichromatic genotypes. MDS results reveal (1) although standardized color perception tests can accurately identify limited color perception abilities (i.e., dichromacy and anomalous trichromacy), these tests can misclassify individuals as color deficient even though the observer demonstrates above average chromatic discrimination; and (2) assessment of the Red to Green region of color space by standardized tests appears to produce ambiguities that lead to misclassifications. Based on these findings we suggest that typically used color vision assessment methods serve as valid tests of dichromacy and anomalous trichromacy, but should not be described as general tests of trichromacy since in some instances the patterns of results observed do not differentiate between observers with superior color vision and dichromat observers. These results are discussed as relevant to models of "normative" color perception, and are related to issues of individual variation in color perception addressed in the symposium.
Color constancy and natural scene statistics: the scope of smart vs. dumb correction algorithms
For natural colors under natural illuminants, the cone excitations for all surfaces in an image are scaled by approximately the same factor with a change in the brighntess or color of the illumination. This allows the effect of varying illumination to be very simply corrected by reciprocal adjustments of sensitivity in the different cone types. The resulting representation is illumination-invariant, but it also fails to preserve information about the overall chromatic cast of a scene. Experimentally this is sometimes inaccurately estimated, with a tendency to "underconstancy" wherein scenes retain in perception a cast of the illuminant color.
"Underconstancy" could result from insufficient sesnsitivity corrections; but when the statistical variation among natural illuminants and scenes is considered, underconstancy can be viewed not as a failure of constancy, but as the adoption of a Bayesian "best guess" about illuminant color, appropriately based on knowledge of relevant environmental statistics. Sophisticated color correction algorithms can also exploit statistics (other than the mean) of the distribution of an image's elements in cone excitation space to resolve the ambiguity inherent in the mean alone. Experiment (Golz and MacLeod) suggests that vision does exploit these cues and gives them statistically justifiable weight.
J. D. Mollon
Coevolution has been defined as 'an evolutionary change in a trait of the individuals in one population in response to a trait of the individuals of a second population, followed by an evolutionary response by the second population to the change in the first' (Janzen, 1980). Could the newer subsystem of primate color vision have coevolved with those tropical trees that present a chromatic signal to their dispersers? Suppose that 30-40 million years ago, occasional sports occurred among the fruits of a given species of tree, sports that were slightly yellower than the foliage in which they were embedded; and suppose that in the same forest there occasionally arose ancestral primates whose vision resembled that of anomalous trichromats, owing to the presence of two slightly different forms of the long-wave retinal photopigment. The yellower fruits would enjoy an advantage by their greater visibility to the anomalous monkey, and the anomalous monkey would enjoy the advantage that came from being able to spot such sports. As the two mutants rose in frequency, we might expect new variants to arise - fruits that were still yellower and retinal photopigments that differed by more than one amino acid. As each trait became more marked, in the plant and in the animal, the advantage of the complementary trait would be enhanced, and so the two advantages would reinforce each other (Mollon, 2000).
It is likely that fruits from different botanical families have evolved to be yellow or orange in color because this signal is salient to trichromatic disseminators. But it is less clear that the trichromacy of primates evolved only because it was of advantage in spotting fruit. Our modelling has shown that the present primate photopigments are optimal for detecting a variety of targets against a foliage background. These targets include not only (i) fruits but also (ii) young leaves, (iii) conspecifics, and (iv) predators of the leopard family (Sumner & Mollon, 200).
Kimberly A. Jameson
Cross-Cultural Color Naming, Color Perception and Cognitive Universals
Current theory on color categorization and naming suggests that pan-human universal neural processes are the basis for similar color naming andcategorization behaviors across cultures. Recent color vision evidence suggests that color processing can substantially differ both intra-culturally and cross-culturally. Recent cognitive research shows that while Basic Color terms are universally the most frequently used to name colors, such terms are always used in the context of larger linguistic systems when describing color experience in some ethnolinguistic groups (for example, monolingual Vietnamese speakers compared to English speakers). Such results illustrate that some populations place a greater reliance upon polylexemic naming, as well as other color-naming constructions which are not easily accommodated by the accepted theory. This result, and other new findings, present new insights and challenges for the search for color naming universals. These challenges will be discussed and in connection with the present symposium topics, and an alternative view relying on pan-human cognitive processing is suggested as the most plausible basis for universal color behaviors.
However cultures presume to categorize, or divide up, the continuously varying chroma within the familiar "color spindle," the assumption has always been that, because we all share the same system of cones and color-opponent cells, all normal humans embody/confront the same color spindle. And the volume within that spindle contains all of the color sensations it is physically possible for normal humans to have. But this last assumption appears to be mistaken. With a little ingenuity, we can indeed produce weird but "recognizable" color sensations that lie outside the volume of the canonical color spindle, but inside the 3-D space of the color-opponent cells. Exploiting a selective pattern of fatigue, across the color-opponent cells in particular, is the key move in a two-step process. It produces 'color sensations' that you have probably never experienced before, or never noticed, if you ever did. These represent 'colors' that do not -- indeed, cannot -- exist in the world of real physical objects. Hence: "chimerical colors." I will try to show them to you.
Constraints on the Development of Color Term Acquisition in Young Children
A question that may provide insight into the brain's cognitive processing of color perception is whether color term acquisition in young children is constrained by any order. Examples of an order might be the one predicted by Berlin & Kay (1969), or an order in which primary color terms appeared before non-primary terms, or an order in which highly visually salient colors appeared before less salient ones. We have tested the development of color conceptualization using two tasks, color naming and color comprehension in 43 children between the ages of 2 and 5 years (Pitchford & Mullen, 2002, Perception). We note that very young children often begin with the inaccurate use of a single color category. Then 9 of the 11 basic color terms all appear rapidly within a narrow time frame of just a few months (yellow, blue, black, green, white, pink, orange, red, purple). We argue that within this frame the acquisition of color terms is largely unconstrained, with no strong evidence that acquisition is driven by the primary or non-primary status of a color, or its visual salience. Interestingly, the final two color terms (brown & grey) appear significantly later (by about 9 months). We therefore believe a two stage process exists with the interior structure of perceptual color space developing later than its exterior. Reasons for this will be speculated upon.
Functions of Colour
Looking through opaque surfaces
What good is transparency perception? How often do we actually have the opportunity to look through one surface and see another? In addition to the standard example of an interposed color filter, surfaces that reflect light specularly provide such an opportunity. There are many situations where one can look through an opaque, shiny surface and see a distorted version of the scene lying in front of the surface, just like a mirror. The standard model for the chromatic combination of diffuse and specular reflection by opaque surfaces works the same way that the standard model of transparency combines chromatic properties. Surface roughness causes the change in contrast associated with the transparency model to depend on spatial variables, so that transparency is involved intimately in perceptual estimates of surface microstructure.
Color and Function
Among the relatively small number of philosophers who have taken in interest in questions concerning the biology and psychology of color vision the prevailing view has been that there is something wrong with any theory of color vision that ties it too tightly to the job of extracting information about the physical properties of distal surfaces. These properties, it is claimed, don't have the appropriate ecological relevance. Some philosophers have argued the contrary, typically claiming that the function of color vision is to extract information about the reflectance of the surfaces in the field of view. Both sides now seem to me to have an uneasy relationship with the empirical literature on color vision and its utility for various organisms. I want (hopefully) to raise three separate issues. First, some organisms, e.g. honeybees, seem to in part to find information about reflectance useful because of local, temporary correlations between the reflectance of surfaces and properties of greater ecological significance such as nectar content. Second, there is a tendency on the part of philosophers to speak of the function of color vision as if there were only a single, significant benefit that affected its evolution. Third, for philosophical purposes it is often useful to talk about the utility of seeing (identifying?) color and thereby ignore the many other uses to which visual systems may put chromatic information.
The function of color: clues from the code
Color gives us a quantitative specification of surface characteristics that can help us recognize things and know their qualities, and also supports spatial vision by enabling detection of object boundaries that provide chromatic contrast. If the former role is primary, the visual system should allocate discriminative ability to different regions in colour space in a way that provides for best discrimination among natural colours. The optimal code in that sense is nearly linear, but not strictly so. Rather, discrimination in an optimized system would satisfy a "cube root rule": differential sensitivity would be greatest for the most commonly encountered colors (near white), dropping to half its maximum under conditions of relative frequency 1/8 (von der Twer and MacLeod).
The fit between this principle and psychophysical observation is improved by considering the stimulus to be the local contrast between test field and background, rather than absolute pixel values. Comparison with physiological data shows less satisfactory agreement: M (achromatic)cells appear to be too nonlinear, and P (chromatic) cells too linear, for forming optimal metric representations of luminance and colour respectively. The all-or-none characteristic of the M cell neural signal is consistent with the idea that the main function of that system is to detect contours, not to represent their contrast quantitatively.
The Dual Nature of `Colour': 'Colour' as Part of the Format of Representational Primitives
The field of colour perception has often been praised in recent years as a paradigm of cognitive science. While this certainly has some validity, it contrasts with the fact that the field makes very little contact with the sort of inquiries into mental representations to be found elsewhere in cognitive science. I find this quite puzzling, because in the earlier literature of the field it was clearly recognised - for instance by Bühler, Gelb, Kardos, or Koffka - that 'colour' could be understood only as part of the general problem of perceptual representations. Their insights could not, of course, take advantage of the theoretical language provided by what has been called the cognitive revolution. For that reason, and also because they were overshadowed by the success of more technical fields, they fell almost entirely into oblivion. The technical fields, successful with respect to their own specific goals, were colorimetry, neurophysiological investigations into peripheral colour coding, and more recently, functionalist-computational approaches that emphasise certain pre-given performance criteria. The success of these fields has not been hampered by the fact that they share certain common-sense conceptions of colour, particularly the idea that colour is an autonomous attribute that can be studied almost in isolation from other perceptual attributes.
I will - from an ethological and internalist perspective (see Mausfeld, in press, for a more detailed account) - provide experimental and theoretical evidence that indicates that 'colour' is not a homogeneous and autonomous attribute but rather plays different roles in different representational primitives. Rather than asking what 'colour' really is, or making presuppositions about its 'proper causal antecedents' or about the 'proper intentional objects' of colour, I will focus on how it figures within the structure of representational primitives of perception. Notwithstanding that we are still far from having a clear theoretical picture about the kind of primitives that underlie perceptual representations, primitives that refer to classes of internal entities such as 'surfaces', 'ambient illumination', '3D-objects', or 'events' (to be understood as internal, and not as physical concepts) suggest themselves as fundamental pillars of the internal representational structure of perception. These primitives determine the data format, as it were, of internal coding. Each primitive has its own proprietary types of parameters, relations and transformations, which define their internal structure and govern its relation to other primitives.
On this account, 'colour' figures as a free parameter in the structure of (at least) two different representational primitives that, from a meta-theoretical perspective, can be regarded as pertaining to the representation of 'surfaces', and the representation of ambient and local 'illuminations'. Consequently, 'colour' does not constitute, as common-sense taxonomies suggest and as most of current research presupposes, a single domain of an autonomous attribute but rather refers to two different types of 'data structure', each with its own proprietary types of parameters and relations. The relation of these representational structures is modulated by classes of parametrised transformations whose effects are mirrored in the idealised computational achievements of illumination invariance of colour codes, on the one hand, and scene invariance, on the other hand. Mausfeld and Andres (2002) found evidence that second-order statistics of chromatic codes of the incoming light array co-determine the decompositions of the retinal colour codes into a dual code and differentially modulate the relation of the two kinds of representational primitives involved.
Uses of color contrast: successes in form, failures in motion
There are two primary uses of color in human vision. One is seeing differences in the visual scene (color contrast) and forms the basis of a figure-ground segregation in space and /or time. The other is the identification of colors and requires color knowledge or cognition. Only the former is relevant to form and motion perception. The question of how, or even whether, we can see form or motion from color contrast has been investigated extensively over the past 25 years. I will mainly discuss form perception. In the early 1980's the 'coloring book' model held that only luminance vision was able to extract contours and edges, with color playing a subordinate role by filling in these luminance defined regions. This model has now been discredited and I will present evidence to show that color vision is able to support visual scene analysis in its own right, exhibiting similar analyzing properties and capacities to luminance vision. This evidence suggests that color contrast is an integral part of the ventral pathways that analyze form, and in this respect cannot be considered as modular. By contrast, psychophysical evidence suggests that color contrast lacks the capacity to support (first order) motion perception, suggesting that the dorsal motion pathway is modular in the sense that it discards chromatic information.
John S. Werner
Utility of Color Vision
Color vision refers to the ability to discriminate spectral distributions irrespective of brightness. It is convenient to discuss color disembodied from other perceptual qualities, but Hering (1920/1964) noted in the last century that colors are always spatial: "Our visual world consists solely of differently formed colors? seen objects, are nothing other than colors of different kinds and forms." This view is supported by modern physiological studies demonstrating that both color and luminance are carried along with form information on the same optic nerve fibers, albeit at different spatial scales. In other words, color vision is inseparable from spatial or form vision. Color and form exist so that we may represent objects, but to be useful this representation must not be based on a one-to-one mapping of the physical world. One reason is that color can vary widely across individual images, and thus objects could not be represented optimally by a fixed visual system. Fortunately, color coding is not fixed, but rather adjusts to both the average color and distribution of colors in scenes through processes of adaptation. Such adjustments may support color constancy and coding efficiency, and may also optimize detection and discrimination of colors that are novel in an image. Finally, the spatial properties of color-coding mechanisms are essential to our perception of figure and ground. Chromatic (border) contrast enhances the difference between figure and ground, while homogenization of object surfaces is facilitated by short- and long-range processes of assimilation and color spreading.