Sparse Representation And Dictionary Learning Based Face Recognition

In pattern recognition,research on face recognition has attracted much attention,and has been successfully applied to many areas of social public security protection.Recently,with the rising of compressive sensing,sparse representation received extensive attention because of its excellent classifying performance and robustness to noise,and it is successfully used in face recognition.Through analyzing and summarizing recent researches about sparse representation-based classification,three improved classification and recognition algorithms are prpposed:1.The LBP feature based on structured sparse representation classification algorithm is proposed.This approach extracts LBP features of the samples and then makes the features input to the structured sparse representation classification algorithm.The algorithm makes full use of the locality of LBP feature and the training dictionary blocking structural of structured sparse representation classification,so it prefers to choose the same class of training samples as much as possible to reconstruct the test samples,and this algorithm greatly improves classification effect.2.The direction Gabor based on kernel sparse representation classification algorithm is prpposed.First of all,the algorithm extracts the Gabor features which are robust to illumination,expression and occlusion.In order to utilize the multi direction feature information of Gabor feature,then makes the direction Gabor features input to the kernel sparse representation classification algorithm based on the coordinate descent method.Therefore this algorithm is robust against illumination,expression and occlusion.3.The competitive agglomeration based on MOD dictionary learning algorithm is proposed.This approach adds clustering algorithm to the classical MOD algorithm,and the clustering algorithm is used to romove the redundant dictionary atoms in the dictionary learning phase to get the better performance dictionares.In order to make the dictionaries discriminative,the MOD algorithm is used to further continue learning.Finally,the dictionaries with superior classification effect is obtained.This paper performs comparison experiments on the AR,ORL and LFW public face databases,in order to verify the feasibility and effectiveness of the proposed algorithms.