Sequential Coordinate-Wise DNMF for Face Recognition

Abstract. This paper proposes the Sequential Coordinate-Wise Algorithm (SCWA) to Discriminant Nonnegative Matrix Factorization (DNMF) for improving face recognition. DNMF incorporates Linear Discriminant Analysis (LDA) into NMF using the multiplicative updating rules that are simple in use but usually require many iterations to converge and they do not guarantee the convergence to a stationary point. The SCWA solves the Quadratic Programming (QP) problem by updating only a single variable at each iterative step, which considerably accelerates the convergence for sequentially projected Least Squares (LS) problems that take place in NMF. Moreover, the limit point of the SCWA is the stationary point. The proposed algorithm is tested for supervised face recognition problems where the facial images are taken from the ORL database.

• Type: Conference
• Year: 2010
• Where: ICAISC
• Authors: Rafal Zdunek, Andrzej Cichocki
• Keywords: Artificial Intelligence, Face Recognition, ICAISC 2010, Multiplicative Updating Rules, Stationary Point