An Approach for the Discrimination of Moving Textures.

The present work aims to bring out a new approach of features selection for the discrimination of textures in sequences of images containing moving objects. The moving textures are analysed using spatio-temporal co-occurrence matrices from which we extract features characterizing the textures themselves as well as their movements. The originality of this approach is to select features that allow differentiating textures either from their visual aspect or from their movements. Our method, based on a supervised learning scheme, uses an iterative selection procedure in order to build a low dimension feature space. It has been applied, tested and validated on two reference databases, namely OuTex and BarkTex.