Quantifying color constancy: evidence for nonlinear processing of cone-specific contrast.

Color constancy was studied by the method of comparing color samples under two different illuminants using a CRT color monitor. In addition to the classical approach in which one of the illuminants is a (standard) white, we performed experiments in which the range of differential illumination was extended by using pairs of lights that were both colored. The stimulus pattern consisted of an array of thirty-five color samples (including five neutral samples) on a white background. A trichromatic illuminant-object interaction was simulated analogous to that resulting from illumination by three monochromatic lights. The test samples, as seen under "test" and "match" illumination, were successively presented to the left and right eye (haploscopic matching). The data show systematic deviations from predictions on the basis of cone-specific normalization procedures like those incorporated in the Retinex algorithm and the von Kries transformation. The results can be described by a nonlinear response transformation that depends on two factors, receptor-specific sample/background contrast and the extent to which the illuminant stimulates the receptor system in question. The latter factor explains the deviations. These are mainly caused by the short-wave-sensitive system, as a consequence of the fact that this system can be more selectively stimulated than the other, spectrally less separated, cone systems.