Unifying Color Constancy

Estimating the color of the illuminant under which an image of a scene was captured is a central part of solving the color constancy problem: that is, of deriving an illuminant independent representation of the re ectances in a scene. In this paper we develop a general correlation framework for solving the illuminant estimation problem. Within this framework estimation is posed as a correlation of the colors in an image with the colors that can occur under each of a set of possible lights. Rather than attempting to recover a single estimate of the illuminant as many previous authors have done, we, in the rst instance, recover a correlation measure for each possible illuminant. We then select an estimate of the scene illuminant based on these correlations. The work presented here follows from previously published work. In this paper we extend that work by showing that the correlation framework is rich enough to allow many existing algorithms to be expressed within it. In particular the grey-world, max-RGB, gamut mapping, and MaloneyWandell algorithms, perhaps the algorithms most widely cited in the literature, are presented in this correlation framework. This work together with work published elsewhere shows that almost all published algorithms based on a Mondriaan world can be formulated in the framework presented here.