Face Recognition Using Improved Extended Sparse Representation Classifier and Feature Descriptor

Representing query samples, many methods based on sparse representation do not take into account the different importance of atoms. In this paper, we propose a new extended sparse weighted representation classifier (ESWRC). In ESWRC, we introduce a representativeness estimator, and use it to estimate the atom representativeness. The atom representativeness is used to construct the weights of atoms. The weighted atoms are used to represent the query samples. In addition, we propose a distinctive feature descriptor, called logarithmic weighted sum (LWS) feature descriptor, which combines the advantages of discrete orthonormal S-transform feature, Gabor feature, co-variance and logarithmic operation. We combine ESWRC and LWS for face recognition and call it improved extended sparse representation classifier and feature descriptor (IESRCFD) method. Experimental results show that IESRCFD outperforms many state-of-the-art methods.