Research On Robust Face Recognition Based On Compressed Sensing

Face recognition technology is a hot research topic in the field of biological identification in recent years and it is widely used in image processing,pattern recognition and other areas of information management because of its non compulsion,secrecy,and efficiency of information acquisition advantages.Face recognition technology has a great role in the national security field.The technology of face recognition with the development of the Internet has become more and more mature,which has promoted a variety of theories to be put forward.Especially the application of compressed sensing tec hnology of face recognition,successfully broke the traditional sampling methods for signal bandwidth constraints.Compressed sensing face recognition technology has some robustness to image recognition with occlusion and illumination.But in practical application,this technique still has some unresolved problems.In this thesis,we focus on the theory of compressed sensing,we improve the algorithm on the basis of the theory.We propose the class mean coefficient variable,fuse sparse coefficients and residuals,redefine the classification strategy and we design a two stage classification strategy,according to the coefficient concentration index.The experimental results show that the proposed algorithm enhances the robustness of face recognition system.The main innovations of this thesis are as follows:(1)We analyze the characteristics of sparse coefficients in the AR and Extended Yale B face database.We propose intra class mean sparse coefficient variable,then verify the correlation between the intra class mean and the residual classification strategy.The intra class mean coefficient effectively weakens the extreme value of the sparse coefficient,and reduces the influence of extreme value on the classification of the residual.This method strengthens the influence of the whole index of sparse coefficient on the classification strategy.(2)Based on the analysis of the correlation between the intra class mean coefficient and the residual,we propose a new method of face recognition based on the fusio n of sparse coefficient and residual named C-SRC.The purpose of this method is to introduce the intra class mean coefficient in the SRC residuals in order to optimize the recognition effect.In this thesis,we define the classification strategy,that is to say,the maximum value of residuals is used to replace the minimum residual value in SRC to determine the category.We use the single sample and multiple samples in the database to verify the method and the results show that the proposed method is robust to the change of expression and illumination.(3)In the classification process,we proposed a two stage classification strategy.According to the sparse concentration index,we propose the index threshold.When the concentration index is lower than the threshold value,we need to choose the five training samples corresponding to the larger residuals to construct a new training sample dictionary,and use the reconstructed dictionary to classify the test samples.When the concentration index is higher than the threshold value,output the result.The method strengthens the verification of abnormal test samples and optimizes the classification strategy.In this thesis,we validate the improved method by adding noise to the image in the face database selected.Compared with the traditional SRC method,the simulation results show that the improved method is superior to the SRC classification.