Research On Face Recognition Method Based On Sparse Representation

At present,with the rapid development of science and technology,personal privacy information leakage incidents occur frequently,causing great damage to people's property and life,biometric identification technology has emerged and developed rapidly.Among them,the face recognition technology not only has high recognition rate,but also will not interfere with the object.Face recognition based on sparse representation has the advantages of high recognition rate,easy to accept,non intrusive,real-time,low cost and so on.Based on the sparse representation of the face recognition algorithm,the test sample can be expressed by linear combination of training dictionary,did not fully consider the representative category of the training sample,and it didn't ignore the training samples which weight to test sample is zero or very small.Aiming at the shortcoming of traditional sparse representation algorithm,in this paper the theory of sparse representation and nearest neighbor classification algorithm combined to obtain the nearest neighbor classification algorithm based on sparse representation and the weighted sparse neighbor representation based classifier for face recognition.The basic theory of nearest neighbor classification algorithm based on sparse representation is: in all training samples,the nearest training samples are obtained by the nearest neighbor classification algorithm and selected to form the training dictionary,then,according to the sparse representation mathematical model to solve the sparse coefficient and the error of each training sample in representing the test samples,the test samples are classified into the categories corresponding to the minimum error.The central idea of face recognition algorithm based on weighted sparse nearest neighbor representation is: first of all,the K training samples nearest to the test samples are selected from each category of training samples and used to form the training dictionary,secondly,before solving sparse coefficient,the weight is appended to the sparse coefficient of each training sample,finally,the training sample of the minimum reconstruction error determines the category of the test sample.In this paper,ORL and Yale face database are used to verify the experiment,and the result show that,the face recognition rate obtained by the nearest neighbor face recognition algorithm based on sparse representation and face recognition algorithm based on weighted sparse nearest neighbor representation was superior to the traditional sparse face recognition algorithm and Principal Component Analysis(PCA),Sparsity Preserving Projections(SPP),Nearest Neighbor Classification Algorithm(NNCM)etc.