Face recognition using region-based nonnegative matrix factorization

Abstract

This paper presents a new method of the face recognition using the nonnegative matrix factorization (NMF) and division of face into several regions. The proposed method divides facial images into 6 sub-regions, and then apply NMF to each sub-region producing basis images and encoding matrices. To recognize a target face, we compare the encoding coefficients of the target image with the encoding coefficients of training images. Test results show that our method is more robust to changes of illumination and facial expression, and occlusions than other methods, and that recognition with 3 sub-regions gives the best result. © 2009 Springer-Verlag Berlin Heidelberg.