A comparison between parts-based and holistic approaches to face recognition

We present a comparative analysis of three distinct parts-based representation approaches to face recognition based on: a) non-negative matrix factorization (NMF); b) local non-negative matrix factorization (LNMF); c) independent component analysis (ICA). Experimental results conducted on AR and Olivetti databases are reported. Results indicate superior performances of specific design solutions over standard eigenfaces approach, although PCA compares favorably to most local methods.