Research On Face Recognition Algorithm Based On Sparse Coding With Local Features

Face recognition algorithm based on sparse representation is quite robust to noise and occlusion. It can maintain a high recognition performance even in the case of a large number of random noise interference or wearing a scarf, glasses and so on. However, in the practical application process, the calculation speed and non-aligned face recognition accuracy are still need to be further improved.In order to solve the existing problems in sparse coding based face recognition algorithm, face recognition technology and sparse coding theory are studied in this paper and the achievements are as follows:This paper proposed an algorithm, the mandatory sparse coding algorithm,which is based both on the accuracy of reconstruction and the computational speed to solve the problem that the high accuracy of reconstruction but the slow computation in sparse representation as well as the fast computation but the low accuracy of reconstruction in collaborative representation. The collaborative representation algorithm is used to generate the collaborative representation coefficient, the result dealt with the mandatory sparse coding algorithm is regarded as the criterion of the initial value and the dictionary dimension reduction in the iterative solution process of sparse representation in order to improve the speed of obtaining the solution of sparse representation with setting the tendentious initial value and reducing the dictionary dimensions.In this paper, we have combined the mandatory sparse coding algorithm with local features, then propose a new face recognition algorithm which is called face recognition algorithm based on mandatory sparse coding with local features. The algorithm use LBP and Gabor as feature extraction algorithm in the stage of construct dictionary so that it improves the recognition accuracy to some extent. Aiming at non-aligned face recognition problem in practical application scenarios,combined with block strategy,face recognition algorithm based on blocked mandatory sparse coding with local features is proposed in order to enhance the recognition accuracy.A large number of comparative experiments on YaleB, AR and other data sets show that the sparsity level of the coding vector which is calculated by mandatory sparse coding algorithm is higher than coding vector that is computed by traditional sparse coding algorithm.Under the premise of ensuring higher recognition rate, the computational efficiency are greatly improved. The computation speed of face recognition algorithm based on mandatory sparse coding with local features is fast and further improve the recognition rate. Furthermore, the accuracy of non-aligned face recognition is greatly enhanced by face recognition algorithm based on blocked mandatory sparse coding with local features which is proposed by this paper.