Research On Occluded Face Recognition Method Based On Sparse Representation And Low Rank Recovery

With the popularity of mobile devices such as digital cameras,smart phones,and intelligent surveillance systems,face recognition technology is gradually being applied in various fields of today's society.However,a series of changes such as illumination,expression,and occlusion in the actual scene make the face information missing,resulting in poor face recognition.This thesis mainly studies the face recognition problem under occlusion,and solves the problem that the recognition rate of the existing method is not high and the running time is long.The specific contents are as follows:Due to the problem that the low rank matrix recovery process takes a long time,this thesis proposes a Fast Low-Rank Matrix Recovery(FLRR)algorithm.This method optimizes the low rank recovery model.In the solution,the matrix with high dimension is divided into the product of three small matrices,which avoids the singular value decomposition of large matrice,reducing the complexity of the algorithm to a certain extent and lays a foundation for the subsequent face recognition process.Due to the face recognition problem of occlusion in training images,this thesis proposes a Collaborative Representation Classification Face Recognition Algorithm Based on Gabor Dictionary Reduction(GDR-CRC).The method performs Gabor transform on the test face image,the low rank face image and the occlusion error image respectively to obtain the corresponding augmented Gabor feature vector.Then,a Gabor occlusion dictionary reduction algorithm is proposed.The reduced Gabor occlusion dictionary is obtained by this algorithm,and it is used together with the augmented Gabor feature vector of the training samples to form a Gabor reduced dictionary.Finally,the test samples are collaboratively represented by a reduced dictionary,and the final recognition result is obtained.The experimental results show that on the basis of guaranteeing the recognition rate of the algorithm,the proposed method reduces the complexity and the running time of the algorithm.Due to the face recognition problem of occlusion in both test images and training images,this thesis proposes a Group Sparse Representation Face Recognition method for Robust Principal Component Analysis(GSR-RPCA).Firstly,transferring the face image from the spatial domain to the logarithmic domain.Recovering each subclass training sample by FLRR algorithm to reduce the correlation between the low rank components and enhance the discriminative ability of recovered data.Then,learning the low rank mapping matrix between the low rank component and the original training data,and using this matrix to map the test sample to its potential subspace to remove the error component.Finally,the group sparse representation coefficient is calculated,and the test face is identified by using the class association reconstruction residual to obtain the category of the test face.Experimental results verify the effectiveness and robustness of this method.