Estimating Scene Illuminant Colors from the Color Images Acquired by a Fixed Camera under Different Illuminations

Abstract

This paper proposes a method for estimating scene illuminant colors from the two color images acquired by a fixed camera under two different illuminations. Our method, which uses the fact that the same scene is observed as the two images, obtains two sets of surface reflectance of a scene area common to the two images, and estimates the colors of the two illuminations based on the property that the intersection of the sets could correspond to the common area. To obtain the sets of surface reflectance from the colors in the images, relationship between surface reflectance and colors in image, which are calculated using surface reflectance called “typical set” from ISO/TR 16066 object colour spectra database for colour reproduction evaluation and possible illuminants, is used. Experiments using numerical simulation and actual images show that our method derives smaller estimation errors than conventional methods and that it provides stable estimations for several kinds of illuminants and reflectance.