The Research Of Fast Face Recognition Based On Sparse Representation

With the improvement of identity authentication security awareness in people's life,using the biometric technology has become one of the preferred methods.The research has become a hot topic at home and abroad.Biometrics has good stability and its own unique nature,while biometric facial features as the more commonly used features,as compared to other features,it's convenient,direct,friendliness andeasy to accept.At present,many algorithms are able to identify and classify the face images accurately,but often due to the high complexity of the algorithm,resulting in time-consuming,can't meet the real-time identification of users' needs.In this paper,starting from the face of sparse representation,the sparse representation principle applied to face recognition,and by building a dictionary,the original data matrix transformation and dimension reduction,thereby avoiding the curse of dimensionality,improving the accuracy of the algorithm and at the same time guarantee the efficiency of the algorithm.Then by weighted training samples,using local information of sample data,and improve the robustness of the algorithm,so as to improve the recognition rate algorithms.The work of this paper mainly includes:First:Based on the traditional SRC algorithm,the paper proposed an algorithm which based on dictionary fast sparse representation classification(DFSRC),using the data dictionary structure of the matrix transformation and using the compressed sensing principle for extracting compressed data matrix,thereby further reducing the dimension of the available data and improving the efficiency of the algorithm to ensure that the ability of real-time recognition.Second:Research on the data locality,comparing with the several different recognition algorithms which use the data local information,and find out the advantages and disadvantages of these algorithms.And then DFSRC algorithm is introduced into data local information and weighted processing,including the distance between the test sample to the various types of training samples and the distance between the test sample to the selected training samples as the weight,thus proposed an algorithm which based on weighted dictionary fast sparse representation(WDFSR),the algorithm ensures efficiency and also improves its classification discriminant ability.Third:In the paper,the traditional SRC,DFSRC and WDFSR algorithm makethe simulation experimentson the face databases(Yale B,ORL and AR database),and compare the recognition accuracy and recognition time of the three algorithms.The results of the experiments showthat the proposed algorithms DFSRC and WDFSR ensure the recognition accuracy while speeding up the rate of implementation of the algorithms,reducing the time required to identify,proved the effectiveness and reliability of the two algorithms.