Color Constancy: Developing Empirical Tests of Computational Models

Brainard et al. 2 INTRODUCTION Object recognition is difficult because there is no simple relation between an object's properties and the retinal image. Where the object is located, how it is oriented, and how it is illuminated also affect the image. Moreover, the relation is under-determined: multiple physical configurations can give rise to the same retinal image. In the case of object color, the spectral power distribution of the light reflected from an object depends not only on the object's intrinsic surface reflectance but also factors extrinsic to the object, such as the illumination. The relation between intrinsic reflectance, extrinsic illumination, and the color signal reflected to the eye is shown schematically in Figure 1. The light incident on a surface is characterized by its spectral power distribution E(λ). A small surface element reflects a fraction of the incident illuminant to the eye. The surface reflectance function S(λ) specifies this fraction as a function of wavelength. The spectrum of the light reaching the eye is called the color signal and is given by C(λ) = E(λ)S(λ). Information about C(λ) is encoded by three classes of cone photoreceptors, the L-, M-, and Scones. The top two patches rendered in Plate 1 illustrate the large effect that a typical change in natural illumination (see Wyszecki and Stiles, 1982) can have on the color signal. This effect might lead us to expect that the color appearance of objects should vary radically, depending as much on the current conditions of illumination as on the object's surface reflectance. Yet the very fact that we can sensibly refer to objects as having a color indicates otherwise. Somehow our visual system stabilizes the color appearance of objects against changes in illumination, a perceptual effect that is referred to as color constancy. Because the illumination is the most salient object-extrinsic factor that affects the color signal, it is natural that emphasis has been placed on understanding how changing the illumination affects object color appearance. In a typical color constancy experiment, the independent variable is the illumination and the dependent variable is a measure of color appearance experiments employ different stimulus configurations and psychophysical tasks, but taken as a whole they support the view that human vision exhibits a reasonable degree of color constancy. Recall that the top two patches of Plate 1 illustrate the limiting case where a single surface reflectance is seen under multiple illuminations. Although this …