The visual system generates object color on the basis of spectral information mediated by reflected light. Although the color of this light, and hence, surface color, varies with the color and intensity of the ambient illumination, the visual system is capable of maintaining a fair degree of color fidelity, a phenomenon known as color constancy.Color constancy was already discussed by Helmholtz, but there still is surprisingly little known about the various underlying mechanisms. The current state of the art is rather confusing, since theoretical and experimental studies tend to follow different routes. Initiated by the work of Buchsbaum, a new class of computational models has emerged, but so far, no attempts have been made to confront these models with experimental data. As if in retribution, no attempts have been made in the recent experimental studies to quantify the data within the context of these new models. As a matter of fact, modeling of color constancy data is hardly ever done. A notable exception is the study by McCann et al. in which a sizeable data set was not only collected, but also theoretically accounted for, the latter within the context of the well-known retinex model.In the following we present some of the results of research that started as an “RGB analogue” of the McCann et al. study, but that eventually turned into a much more extensive PhD project. Here only two of the variables tested will be discussed: the chromaticity and the spectral composition of the illuminant. The data we obtained not only allowed the derivation of a simple (but accurate) quantitative model, but also provided the first experimental test of the basic principles underlying the aforementioned computational approaches to color constancy.
Jan Walraven, Marcel P. Lucassen, "Quantifying Human Color Constancy" in Proc. IS&T 1st Color and Imaging Conf., 1993, pp 60 - 64, https://doi.org/10.2352/CIC.1993.1.1.art00014