Object color constancy for outdoor multiple light sources

Abstract

Color is one of the mostly applied features for object recognition and tracking. Most work for color constancy is often based on the assumption of spatial uniformity or smooth illuminant transaction, which is not always true due to the presence of multiple light sources. In this paper, without these assumptions, we deal with the problem of color constancy in multiple light sources by computing the color constancy on a given object rather than the whole image. It keeps the color constancy for a given object under different outdoor lighting conditions, especially for an object under different shadows. We first calculate a transfer vector based on the given object and the illuminants ratio vector. This vector is then added to the original image to make the object be perpendicular to the illuminants ratio vector. Finally, an object color constant image is obtained by performing an orthogonal decomposition along the illuminants ratio vector on the new image. Compared with color constancy on whole image, this proposed method can reduce color distortion in the object and keep mostly color constancy for an object to be recognized and tracked regardless of lighting conditions. Both quantitative and qualitative experiments validate our method.