Face Recognition Based On Feature Fusion For Low Rank Recovery Sparse Representation

In recent years,face recognition has become one of the important research fields on computer vision and pattern recognition because of its broad application prospects.With the introduction of compressed perceptual coding theory,face recognition technology based on sparse representation of classification(SRC)has attracted the attention of many researchers.In SRC,the test image is coded as a sparse linear combination of training samples,and then the solution is obtained by solving the norm optimization problem,which has achieved good effect in practical application.However,the unit matrix is used as the error matrix in SRC,the description of the error and noise in the sample is not accurate.Therefore,this paper introduces the low rank recovery algorithm to separate the error matrix,and decompose the sample image into a clean low rank matrix and sparse error matrix,so that the image information for classification and recognition is more effective.Considering that face images are influenced by emotions and gestures,light and noise and so on,the error rate will be high.Therefore,many scholars have begun to integrate various features into face recognition.In order to further improve the recognition accuracy of face recognition in complex environment,this paper proposes a method based on feature fusion of low rank recovery sparse representation.Firstly,the low face recovery algorithm is used to obtain the clean face images of the training samples and the test samples,and the LBP,HOG,Gabor vectors are extracted from these clean face images.In the training stage,some training samples are randomly selected for SRC classification test,and a loss function is defined by the recognition result of SRC and the classification residual,then use the least squares method to calculate the weight vector which minimizes the loss function.In the recognition stage,the final residuals of the test samples are recalculated according to the weight vector calculated in the training stage,and the final recognition result will be obtained.The results show that this method is superior to using only a single feature recognition method and is robust to light,noise,occlusion,and so on.The research work of this paper mainly includes the following aspects:(1)This paper summarizes the background and research status of face recognition,introduces the classification of face recognition methods,summarizes the key problems in face recognition,and introduces the theoretical problems of image preprocessing,feature extraction and dimension reduction.(2)This paper describes the sparse representation of the face recognition method.Firstly,the theory of sparse representation is introduced.Secondly,the process of face recognition based on sparse representation is discussed.Finally,the samples with different feature dimensions are tested and compared.(3)It is not accurate to describe the error matrix by using the unit matrix in the sparse representation model.The low rank recovery algorithm is used to deal with the samples,which can effectively isolate the influence of the external environment factors on the samples and solve the incomplete dictionary problem caused by less training samples.(4)A new method of face recognition based on feature fusion is proposed to solve the problem that the singularity of a single feature is limited.In the training stage,a loss function is defined according to the classification result of SRC and the classification residual,and the weight vector with the smallest loss function is calculated by using the least squares method.In the recognition stage,the recognition result is calculated by the residual of test sample,which calculated according to the weight vector..(5)Compared to three single feature recognition algorithm,the proposed method was validated on different face database by a wide range of experiments.