2009 | OriginalPaper | Chapter
Localized versus Locality Preserving Representation Methods in Face Recognition Tasks
Four different localized representation methods and two manifold learning procedures are compared in terms of recognition accuracy for several face processing tasks. The techniques under investigation are: a) Non-negative Matrix Factorization (NMF); b) Local Non-negative Matrix Factorization (LNMF); c) Independent Components Analysis (ICA); d) NMF with sparse constraints (NMFsc); e) Locality Preserving Projections (Laplacianfaces); and f) Orthogonal Projection Reduction by Affinity (OPRA). A systematic comparative analysis is conducted in terms of distance metric used, number of selected features, and sources of variability on AR, Yale, and Olivetti face databases. Results indicate that the relative performance ranking of the methods is highly task dependent, and varies significantly upon the distance metric used.
