An Extended Max-margin Non-negative Matrix Factorization for Face Recognition

Abstract

Non-negative matrix factorization (NMF) is a dimension-reduction technique based on a low-rank approximation of the feature space. Unfortunately, most existing NMF based methods are not ready for encoding higher-order data information and ignore the local geometric structure contained in the data set. Additionally, the previous classification approaches which the classification and matrix factorization steps are separated independently. The first one performs data transformation and the second one classifies the transformed data using classification methods as support vector machine (SVM). In this paper, therefore, we joint SVM and constrained NMF into one by uniting maximum margin classification constraints into the constrained NMF optimization. Experimental results on the benchmark image datasets demonstrate the effectiveness of the proposed method