Feature extraction by combining independent subspaces analysis and copula techniques

A method using copula techniques to capture the dependence structure inside the independent feature subspaces is proposed in this paper. It differs from the previous approach that simply use the norm of the projection of visual data on the invariant feature subspace to give the probability density inside the independent subspaces. By modelling the independent feature subspaces with Archimedean copula and utilizing the relationship between Archimedean copula and ℓ₁-norm symmetric distribution, we make use of the corresponding radial distribution as the feature information to process feature extraction.