Previous: Problem Definition Up: Problem Definition Next: Color Naming

Color Perception

As John Locke observed, objects do not have colors [Locke 1690]. Color is not a physical phenomenon, but a perceptual phenomenon that is related in a complex way to the spectral characteristics of electro-magnetic radiation in the visible wavelengths striking the retina [Ronchi 1957][Danger 1987][Boynton 1990][Wyszecki \& Stiles 1982]. An illustration of the perceptual nature of colors is the phenomenon of metamers, or spectrally different stimuli that are indistinguishable to human observers, and hence will be perceived as the same color. For example, Judd has pointed out that the color of a Bunsen burner flame into which sodium is introduced is very similar to the color of an orange (the fruit) in daylight, yet their spectra are almost perfectly complementary [McIlwain \& Dean 1956][p. 24]. In fact there are infinitely many different spectral stimuli that will be perceived as the same. In the discussion that follows I roughly adhere to a division into the three domains of physics, physiology, and psychology that are involved in optical phenomena, following [Ronchi 1957].

The physical stimulus involved in color perception is light, or more precisely: electro-magnetic radiation in the visible wavelength range, approximately 380-770 nm. I will represent the spectral energy distribution of the stimulus (the radiation striking the retina) as a function of wavelength after [Boynton 1990]. The stimulus energy distribution of the radiation that is reflected off (or transmitted by) an object is determined by (1) the spectral distribution of the radiation incident on the object (the light source) , and (2) the spectral reflectance characteristics of the object , so we can write

in a simplified form. It is easily seen from this equation that changing the light source or changing the reflectance characteristics of the object changes the stimulus . Were we informally to think of object color as , then the task of the color vision system is to recover from (for some examples of this approach, see [Maloney 1993][Maloney \& Wandell 1986]). Or as [Zeki 1993] puts it:

The brain strives to acquire a knowledge about the permanent, invariant and unchanging properties of objects and surfaces in our visual world. But the acquisition of that knowledge is no easy matter because the visual world is in a continual state of change. Thus, the brain can only acquire knowledge about the invariant properties of objects and surfaces if it is able to discard the continually changing information reaching it from the visual environment. [p. 355]

The human visual system is to some extent able to identify the same ``object color'' (surface reflectance, under the view just mentioned) under widely varying lighting conditions, a phenomenon known as color constancy, but I will not be concerned with that. Color constancy introduces considerable complexity that is not central to the aims of my work. What I am concerned with is the internal structure of the color categories that invariant object colors are perceived as belonging to. That is not to say that I will adhere to the ``color science'' aperture mode of color perception either; I will take a psycho-physical approach to modeling color perception, while at the same time keeping neurophysiological findings in mind.

The response of the visual system to brings us into the domain of neurophysiology. The human visual system is sensitive to differences in due to the presence of three photo-pigments with different spectral sensitivities in three types of photoreceptor cells (called cones) in the retina [Leibovic 1990b]. The cone types are sometimes called red, green, and blue, for the spectral color corresponding approximately to their wavelength of maximum sensitivity, but more appropriate designations are long, medium, and short wavelength sensitive cones, respectively. If we represent the cone action spectra (or linear transforms thereof, see [Horn 1986]) as and integrate them with the spectral energy distribution of the stimulus [Horn 1986][Boynton 1990], we obtain the so-called CIE tristimulus values :

Any two stimuli that result in identical values cannot be distinguished from each other; i.e. they are metamers.

So far I have only described the response of the cone type photoreceptors to . A lot more is involved in the physiology of color perception, and below I will refer to subsequent stages of processing as necessary.

From the psychological (or psychophysical) point of view, there is a lot more to be said about color perception as well, see e.g. [Boynton 1979][Boynton 1990]. Some important psychophysical dimensions of color perception are brightness or lightness (how bright a visual stimulus appears to be, viewed in isolation or in context), chromaticity or hue (what enables us to distinguish between equally bright and texturally identical fields - this is the closest to the intuitive concept of color), and saturation (how pure a perceived color is, or how unlike grey). Another important concept is chromatic context, or the finding that the appearance of a color depends importantly on its surroundings. Also, the intensity of the light source, and even memory and psychological context can affect color appearance [Boynton 1979]. I will return to some of these issues below.

lammens@cs.buffalo.edu