Reserch On The Application Of Sparse Representation And Seif-quotient Image In Face Recognition

Face recognition,as one of the important research topics in the field of computer vision and pattern recognition,has very wide application value.Face recognition technology has developed rapidly in recent years,the sparse representation classification(SRC)as an emerging field,has been successfully applied to face recognition because of its many advantages such as simple form,high reconstructed signal accuracy,and stronger robustness,and achieved a very good recognition effect and robustness.However,most of the current sparse representation classification algorithms are based on global features,which makes it hard to overcome the effects of a series of changes in occlusion,pose,expression,and illumination in the face image,and the low real-time performance restricts the SRC application in the actual production and life.Aiming at the shortcomings of SRC algorithm,this paper deeply studied the local structure information of data and introduced to the SRC model,aiming to improve the accuracy and robustness of the recognition algorithm.The main research contents are as follows:1.Aiming at the problem of insufficient performance of the SRC algorithm under large intra-class changes,an improved weighted sparse representation classification algorithm(WSRC_DALM)was proposed.The algorithm used the gaussian kernel distance to measure the similarity between the training sample and the test sample,and improved the gaussian kernel parameters,and the image was processed in blocks to realize feature dimensionality reduction.And then the dual augmented lagrange multiplier(DALM)method was used to solve sparse representation coefficients and reconstruct test images.Finally,the classification was determined by reconstructing the residuals.2.In the view of the problem of face recognition under complex illumination changes,the face recognition method combining self quotient image(SQI)and WSRC_DALM algorithm was presented,by using the self-quotient image algorithm to eliminate the influence of illumination,the face images under standard lighting conditions was obtained,and then WSRC DALM was used to identification.The proposed face recognition method at the same time can solve the problem of face recognition with large intra-class changes and complex illumination changes.3.With the ORL and FEI face databases,the recognition rate and time efficiency of the traditional sparse representation classification algorithm(SRC),weighted sparse representation classification algorithm(WSRC)and WSRC_DALM algorithm was compared;and the face recognition method based on self-quotient image and WSRC DALM was verified on Yale B and CMU-PIE face databases.The experimental results indicate that:(1)The improved weighted sparse representation classification algorithm can effectively improve the recognition rate and robustness of face recognition in larger intra-class changes(multi-pose);(2)The self-quotient image method can effectively eliminate the influence of illumination of face image;(3)The combination of the self-quotient image and WSRC DALM algorithm can not only improve the face recognition rate due to illumination,but also can enhance the robustness of the algorithm for face images with mixed effects of lighting,pose,expression and so on.