A Computational Model for Color Constancy by Separating Reflectance and Illuminant Edges within a Scene.

Abstract

Several psychophysical experiments have revealed that color constancy exists even in shadowed scenes where the spatial power distribution of the illuminant has abrupt step changes, while previous models for color constancy fail to distinguish shadows and reflectance changes. This article addresses this problem and proposes a computational model for color constancy which separates reflectance and illuminant edges by two modules recovering both the spatial power distribution of the illuminant and surface reflectances of an object from an input scene. Each module has a line process which interprets whether a given edge is derived from a reflectance or illuminant change based on prior knowledge of shadows: it is mainly luminance that changes at the boundry of a shadowed area. Simulation results show that the proposed model correctly detects reflectance and illuminant edges and illuminant colors, and successfully removes the illuminant from the input scene. This suggests that recovering both surface reflectance and the spatial power distribution of the illuminant is one of the possible ways to achieve color constancy in shadows. Copyright 1996 Elsevier Science Ltd.