Towards illuminant invariant texture classification

This paper provides a short overview of work recently performed by the author into the effects of illuminant variation on texture classification, and proposes one method to reduce misclassifications caused by such variations. The paper first presents a model of image texture originally introduced by Kube and Pentland and further developed by the author. The effect of variation in the tilt angle of the illuminant is high-lighted. A set of tilt-compensation filters is developed from the model. The filters are used to pre-process texture images used in training and classification sessions