Cost sensitive matrix factorization for face recognition

Abstract

In this paper, we propose a cost sensitive matrix factorization (CSMF) for face recognition. To make the face representation cost sensitive, CSMF adopts a more flexible feature embedding strategy. It contains two main steps: (1) matrix factorization for the learning of latent semantic representation and (2) cost sensitive latent semantic regression. In this way, the face images are embedded into their label space with the misclassification loss minimized. The experimental results on Extended Yale B and ORL demonstrate its effectiveness.