MBS 93-36
Probabilistic Color Constancy
Michael D'Zmura, Geoffrey Iverson, Benjamin Singer
Specifying the frequency with which surface reflectance functions occur
in the visual environment lets one use the chromaticities of reflected
lights to provide maximum likelihood estimates of the spectral properties
of a scene's illuminant. this approach to color constancy generalizes,
in a natural way, schemes that use the gray-world assumption. Monte Carlo
simulation suggests that a trichromatic visual system needs the chromaticities
of reflected lights from a random sample of at least four surfaces to estimate
accurately the correlated color temperature of daylight illumination.