Abstract
In order to perceive surface colors as more constant under changes of illumination than could be expected on the basis of the illumination-dependent retinal input, the visual system has to decompose the retinal input into two representations, an estimate of the illuminant and an illumination-independent designator for the color of the surface. I will consider one particular class of cues that the visual system could use for estimating the chromatic properties of the illuminant: statistics of the illumination-dependent chromatic distribution within the retinal image. The approach presented uses two heuristics to derive potentially useful candidate cues (an idealized model of our chromatic environment and a simulation with images of natural scenes) and psychophysical experiments to test whether the human visual system exploits these candidate cues. Using these heuristics, the correlation between chromaticity and luminance within the retinal image has been found to be diagnostic of the chromatic properties of the illuminant and the experimental results indicate that this statistic is taken into account by the human visual system in order to estimate the chromaticity of the illuminant.