Abstract—Accounting color information in the texture description can help in the context of classification but the instability of color across illumination variations remains a problem since most practical applications are more likely to face varying illumination conditions. That is why we analyze the impact of such variations on the most classical texture features and we show that, from their structures, these features are not far from the lowest invariance degree required by most of the applications. Hence, applying classical color normalization on these features leads to unnecessary invariance that tends to decrease their discriminative power. Consequently, in this paper, we propose a feature transformation model and a deduced normalization step. We show that the resulted texture features do not depend on the illumination variations while preserving the maximum of the discriminative information. Our classification results outperform the existing texture descriptors available in the literature.
Index Terms—Texture, color, illumination invariance, classification.
The authors are from the Laboratoire Hubert Curien UMR CNRS 5516 Université Jean Monnet – Saint Etienne, France 42000.
Cite: Rahat Khan, Damien Muselet, and Alain Trémeau, "Texture Classification across Illumination Color Variations," International Journal of Computer Theory and Engineering vol. 5, no. 1, pp. 65-70, 2013.