A New Face Recognition Framework Based on Improved Nonnegative Matrix Factorization

Abstract

A new face recognition framework based on improved Nonnegative Matrix Factorization (NMF) is proposed in this paper. The improved NMF algorithm adds a sparse constraint which makes the partbased features of images more representative. We develop a novel multiplicative update rule for the algorithm and prove its convergence by using an auxiliary function method. The framework of the face recognition system has three main steps. Firstly, apply the two-level Haar wavelet decomposition to transform images to a low-dimensionality space. Secondly, use the improved NMF algorithm for feature selection. Finally, adopt Support Vector Machine (SVM) for classification. Experiments contrasted with traditional algorithms demonstrate that the proposed method has high classification accuracy with high processing speed. Experimental results also express that the dimensionality of feature subspace is able to affect face recognition accuracy.