The Research Of Face Recognition Based On Sparse Representation

With the rapid development of computer, information security is more and more important in human daily life. The face recognition based on sparse representation was widely developed because of its strong robustness. On one hand the sparse representation of face basically adopt an orthogonal matching pursuit algorithm from greedy algorithm to solve the sparse coding coefficient, leading to the insufficient precision of sparse coding coefficient and less accurate face reconstruction. On the other hand original dictionary is adopted in SRC based face recognition, if each type of face training set is too big, it will make the dictionary become too large and enlarge calculation. This paper deeply studied the solving of sparse coding and dictionary learning which based on sparse representation, and puts forward three effective methods in face recognition based on sparse representation:(1) Face recognition based on subspace pursuit. We put forward an algorithm about face recognition based on subspace pursuit. Iterative optimization idea is used in the algorithm. The number of elected atoms is determined according to sparse degree and updated using optimization to better composite the original signal which have better robustness and expressing ability to reconstruct face to improve face recognition performance. Experimental results showed that the method were superior to the traditional sparse representation in the recognition rate, recognition time and robustness.(2) This paper proposes a face recognition method based on improved subspace pursuit in view of that the sparse degree of subspace pursuit algorithm needs to be known in advance. Our method converges the sparseness, gradually increase the number of added and removed atoms in each iteration in candidate set to approximate the sparse degree. The method is proved to be a better performance when compared with other methods in the face recognition.(3) Face recognition based on improved sparse representation model and LC-KSVD algorithm is proposed. There are a lot of occlusion and damages in face images, and with a lot of noise. To deal with this problem, this paper puts forward an improved sparse representation model. Items of the noise and damage are added in expression to increase the face representation ability of a dictionary which keeps out the influence of noise and damages. LC-KSVD technology is introduced to reduce the complexity of the algorithm. Experimental results show that when compared with the traditional sparse representation of face recognition our method has a huge advantage in terms of recognition speed and recognition rate.