Face recognition algorithm based on improved kernel sparse representation

Abstract

When the face recognition technology captures the sample image, the recognition performance is rapidly reduced due to the change of the shooting angle and distance, the different illumination brightness, and the variability of the facial posture. In order to solve the problem of insufficient computational complexity of SRC algorithm and insufficient number of training samples, this paper proposes a face recognition algorithm based on discriminative low rank decomposition and fast kernel sparse representation. On the basis of low rank decomposition, the discriminability of the low rank matrix is improved, and the discriminative low rank approximation matrix and the sparse error matrix are combined to form a dictionary for testing. In addition, the kernel coordinate descent method is used to solve the sparse representation matrix. The experimental results on the AR face database show that the proposed algorithm guarantees the recognition accuracy and improves the recognition efficiency when the training samples are insufficient.