Computational Models of Color Constancy

For two different purposes we want to study color constancy. First, it is always of interest to know how the biological visual system achieves some relative color constancy. Second, it is attractive to build some artificial visual system that exhibits color constancy for possible practical applications.

To account for the color constancy, various computational mechanisms have been proposed. The ultimate goal of all of the mechanisms is to separate and suppress the illumination $I(\lambda)$ in the light $L(\lambda)$ received by the retina cone cells, so that the reflection $R(\lambda)$ independent of the illumination can be recovered. The key to this separation is the assumption that the illumination is relatively homogeneous while the reflectance may contain more abrupt spatial changes.