| By virtue of its stability, friendly, concealment and other advantages, face recognition technology is receiving more and more attention in the field of biometrics recognition. In the ideal and a better environment for simple cases, identification of many existing face recognition algorithm can have a higher rate, but when the environment becomes more complex, appear to pose and illumination and occlusion problem, the existing algorithms of recognition rate will be greatly decreased, the accuracy of face recognition cannot be guaranteed, the application of the algorithm is also difficult. In this paper, to light the real-time face recognition algorithm under the condition of the research, through the simulation of the initial visual cortex receptive field, focusing on the light decomposition model of non-negative matrix with sparse constraint according to processing algorithm, and carried out some research work on the subject. The main work of this paper include the following:1, on the face recognition and light treatment of research and analysis, research on the sparse vector method, the Euclidean distance from the closest vector sparse algorithm,proposed a vector sparse algorithm, Manhattan distance to the closest experimental results show that the proposed algorithm for sparse vector sparse speed, good effect.2, the non-negative matrix factorization algorithm is studied in this paper, the sparse constraint and non-negative matrix factorization algorithm combined with sparse constraint of non-negative matrix factorization algorithm based on 3 kinds of decomposition algorithm are compared, pointed out the advantages and disadvantages of 3 kinds of decomposition algorithm, the experiments show that only in the basis matrix W or only in coefficient H matrix sparsity constraint, the reconstruction error is small.3, research on the NMF based illumination normalization algorithm, the standard of illumination... |
PDF Full Text Request