Research On Face Recognition Algorithm Based On Sparse Representation And Support Vector Machine

Face recognition technology is a recognition technology based on physiological characteristics in biometrics field.Because of its easy operation and good sealing property,it is widely applied in intelligent monitoring,security system,criminal investigation and so on.In this paper,two kinds of recognition methods are studied,namely face recognition based on sparse representation-based classification(SRC)and face recognition based on support vector machine(SVM).The following research works are carried out in this paper:(1)The paper proposes a new face recognition algorithm named LBP dictionary sparse based on low rank subspace.Firstly it uses LR algorithm to decompose low rank matrix of training samples,and then extract the LBP features of low rank structure and test samples,next gets feature dictionary and the test sample feature by PCA dimensionality reduction LBP,then construct the sparse representation classification model for the classification and recognition.Contrast experiments on YaleB face database showe that the proposed method has better robustness and recognition efficiency.(2)The paper puts forward a face recognition algorithm based on weighted Gabor feature and SVM.Firstly,the images are uniformly non-overlappingly partitioned,weighted according to the image entropy of each sub-block,and then the Gabor features of each sub-block are extracted.After proper dimension reduction,the Gabor features of all sub-blocks are combined into a new Gabor feature Finally,using support vector machines for classification and identification.Experimental results on Yale face database show that the proposed method has better recognition rate than other methods,and it reflects the robustness and robustness in the low dimensional case.(3)The paper gives a new algorithm of face recognition based on KPCA dictionary sparse and SVM.The algorithm is proposed on the basis of the above two work,which combines SRC and SVM to improve the recognition efficiency.Firstly,KPCA features of samples were extracted and KPCA feature dictionary was constructed.All training samples were sparsely represented by feature dictionary to obtain training set data.Then the test samples were sparsely obtained to obtain test data,and then classified and identified by SVM.Compared with the traditional SRC method and SVM method,experimental results proves that the proposed method has certain superiority.