Dissimilarity computation through low rank corrections

Most of the energy of a multivariate feature is often contained in a low dimensional subspace. We exploit this property for the efficient computation of a dissimilarity measure between features using an approximation of the Bhattacharyya distance. We show that for normally distributed features the Bhattacharyya distance is a particular case of the Jensen-Shannon divergence, and thus evaluation of this distance is equivalent to a statistical test about the similarity of the two populations. The accuracy of the proposed approximation is tested for the task of texture retrieval.