Abstract
Surfaces returning identical light spectra to the eye can elicit different color percepts when embedded in spectrally different surrounds. Although various theories have been put forward to rationalize these color contrast effects, there is no consensus about their basis. Here we tested the hypothesis that color contrast is generated by the probability distribution of the possible physical sources of spectral stimuli (see Lotto and Purves, PNAS 97:12834, 2001). The analysis used a database of 41 natural scenes in which the radiance spectrum of each point (i.e., the projected light spectrum for each pixel in the images) was known. The relevant reflectance spectra for corresponding points were computed by removing the influence of both the illuminant and scene geometry from the radiance spectrum. The illumination spectra for each pixel (which include the influence of both the illuminant and scene geometry) were then determined by dividing the radiance spectrum by reflectance spectrum. To facilitate the statistical analysis, the radiance spectrum on each pixel was converted into RGB tristimulus values. Each image in the database was sampled repeatedly with a center/surround template, and the probability distributions of the possible combinations of reflectance and illumination spectra that could have generated the relevant RGB values were determined. The probability distributions of the reflectance and illumination spectra of the central target varied as a function of the RGB values of the surround, indicating that the typical physical sources of target spectra differ when they are embedded in spectrally different surrounds. The color percepts predicted by these distributions were in good agreement with the percepts elicited by color contrast stimuli. This evidence supports the conclusion that color contrast effects are determined by the probabilistic relationship between ambiguous spectral stimuli and the distribution of their possible sources.