Weighted collaborative representation and classification of images

Recently a collaborative representation (CR) based classification with regularized least squares (CRC-RLS) has been proposed for the classification of faces. CRC-RLS is a simple yet fast alternative to sparse representation (SR) based classification (SRC). While SR is the solution to an l₁-regularized least square decomposition, CR starts from an l₂-regularized least square formulation. In this paper we extend the CRC-RLS approach to the case where the samples are weighted based on classification confidence and/or the feature channels are weighted using variance. The weighted collaborative representation classifier (WCRC) improves classification performance over that of the original formulation, while keeping the simplicity and the speed of the original CRC-RLS formulation.