Separability-Based Kullback Divergence Weighting and Filter Selection for Texture Classification and Segmentation

Features computed as statistics (e.g., histograms) of local filter responses have been reported as the best descriptors for texture classification and segmentation. The selection of the filter bank remains however a crucial issue, as well as exploiting a relevant combination of these descriptors. Here, we propose a novel approach relying on the definition of the texture-based similarity measure as a weighted sum of the Kullback-Leibler measures between feature statistics. The weights are computed according to the maximization of a margin based criterion. This weighting scheme can also be considered as a fast filter selection method: texture filter response distributions are ranked so that the problem of selecting a subset of filters reduces to picking the first features only. Experiments carried out on Brodatz textures and sonar images show that the proposed weighting method improves the classification rates while considerably reducing the number of the features. 1