Performance analysis of different matrix decomposition methods on face recognition

Applications using face biometric are ubiquitous in various domains. We propose an efficient method using Discrete Wavelet Transform (DWT), Extended Directional Binary codes (EDBC), three matrix decompositions and Singular Value Decomposition (SVD) for face recognition. The combined effect of Schur, Hessenberg and QR matrix decompositions are utilized with existing algorithm. The discrimination power between two different persons is justified using Average Overall Deviation (AOD) parameter. Fused EDBC and SVD features are considered for performance calculation. City-block and Euclidean Distance (ED) measure is used for matching. Performance is improved on YALE, GTAV and ORL face databases compared with existing methods.