IDENTIFICATION OF ILLUMINANT AND OBJECT COLORS - HEURISTIC-BASED ALGORITHMS

In everyday scenes, from perceived colors of objects and terrains, obs ervers can simultaneously identify objects across illuminants and iden tify the nature of the light, e.g., as sunlight or cloudy. As a formal problem, identifying objects and illuminants from the color informati on provided by sensor responses is underdetermined. It is shown how th e problem can be simplified considerably by the empirical result that chromaticities of sets of objects under one illuminant are approximate ly affine transformations of the chromaticities under spectrally diffe rent illuminants. Algorithms that use the affine nature of the correla tion as a heuristic can identify objects of identical spectral reflect ance across scenes lit simultaneously or successively by different ill uminants. The relative chromaticities of the illuminants are estimated as part of the computation. Because information about objects and ill uminants is useful in many different tasks, it would be more advantage ous for the visual system to use such algorithms to extract both sorts of information from retinal signals than to discount either automatic ally at an early neural stage. (C) 1998 Optical Society of America.