Occluded face recognition using parts-based representation methods

Occluded face recognition performances of four different parts-based representation approaches are presented. The methods under investigation are: a) nonnegative matrix factorization (NMF); b) local NMF (LNMF); c) independent components analysis (ICA); d) neural autoassociators (NA). A systematic comparative analysis is conducted in terms of distance metric used and number of selected features on AR and Olivetti databases. Simulation results indicate superior performances of specific design solutions over classical holistic approaches based on principal components analysis (PCA).