Optimal spatial filter selection for illumination-invariant color texture discrimination

Color textures contain a large amount of spectral and spatial structure tha t can be exploited for recognition. Recent work has demonstrated that spati al filters offer a convenient means of extracting illumination-invariant sp atial information from a color image, In this paper, we address the problem of deriving optimal filters for illumination-invariant color texture discr imination. Color textures are represented by a set of illumination-invarian t features that characterize the color distribution of a filtered image reg ion. Similar features have been used in previous studies. Given a pair of c olor textures, we derive a spatial filter that maximizes the distance betwe en these textures in feature space. We provide a method for using the pairw ise result to obtain a filter that maximizes discriminability among multipl e classes. A set of experiments on a database of deterministic and random c olor textures obtained under different illumination conditions demonstrates the improved discriminatory power achieved by using an optimized filter.