Face Recognition Based on Dictionary Learning with the Locality Constraints of Atoms

Previous dictionary learning algorithms usually take the locality information of training samples into account in the learning process, and it may degrade the robustness of the dictionary. In this paper, an new locality constrained dictionary learning algorithm (LCDL) for face recognition by using the locality characters of atoms is proposed. Since the atoms are learned from the training samples, they are more robust to the noise and outliers than training samples. In the LCDL algorithm, we use atoms to construct a Laplacian graph, and then use the profile (the row vector of coding coefficients matrix) to measure the similarity among them. Then, we construct a locality constraint term by using the profile matrix and Laplacian graph of atoms. Since the profile and atoms can be adaptively updated in dictionary learning processing, the locality constraint term also can be adaptively updated. Moreover, the locality constraint term also can inherit the geometrical structure of the training samples, and it can enhance discriminative ability of the dictionary. Experiment results show that the LCDL algorithm achieve more promising performance than some state-of-the-art dictionary learning and sparse coding algorithms.