IR and visible face identification via sparse representation

We present a face recognition technique based on the sparsity principle. Parsimony is used both to compute the face feature vectors and to process the classification of these vectors. Applied to visible and infrared modalities on the Notre-Dame database, we show that this approach has equal or better performances than those of the state-of-art on this database. This classification allows to use a simple method to merge the scores of these two modalities in order to enhance significantly the identification rates. We show also that this approach is quite robust to corrupted probe images.