Research On Face Recognition Algorithm Based On Sparse Representation

The face recognition technology is a kind of identity recognition method that uses the computer to analyze the face images and extract the effective feature to discriminate identity.When the face image is affected by factors such as expression,illumination,posture,and occlusion,the traditional face recognition technology cannot quickly and effectively recognize the face image.Sparse representation is one of the most popular theories in the field of face recognition in recent years.This theory has certain robustness to illumination,noise and occlusion when the number of samples is large.In order to improve the robustness of sparse representation theory in the absence of training samples,this paper improves sparse representation theory.The main research work is as follows:First,the research history and current status of face recognition algorithms at home and abroad are summarized.In order to reduce the computational complexity of the sparse representation theory,this paper analyzes and studies the method of extracting facial image features.Principal component analysis,linear discriminant analysis and kernel principal component analysis algorithm are used to extract facial image features to reduce image dimension.The research laying the foundation for the subsequent recognition of face images.Second,a face recognition algorithm based on discriminative low rank decomposition combined with fast sparse representation is proposed.The algorithm combines discriminative low rank matrix and error matrix into the dictionary used in the test,and then uses the coordinate descent method to solve the sparse representation matrix.This algorithm solves the problem that the Sparse Representation-based Classifier algorithm can reduce the recognition accuracy when the number of training samples is insufficient and the complexity of solving the₁l norm optimization problem.Different comparative experiments were performed on the YALE and ORL face database,confirming the effectiveness and enforceability of the improvement.Thirdly,a face recognition algorithm based on fast kernel extended sparse representation is proposed.The method of the face recognition algorithm based on fast kernel expansion sparse representation is proposed.It used the sample itself and the intraclass variables to form a dictionary,and then used the nuclear coordinate descent method to solve the problem of₁l norm optimization in the kernel space.This algorithm makes full use of the nonlinear information of the original data samples,and reconstructs the test samples by using the training sample feature information in the kernel space.The experimental results on the AR and YALEB face databases show that the proposed algorithm improves the recognition efficiency and guarantees the recognition accuracy under the condition of insufficient training samples.Finally,a face recognition algorithm based on discriminative low rank decomposition and fast kernel sparse representation is proposed.This method the basis of low-rank decomposition,improve the discriminability of the low-rank matrix,combining the obtained discriminative low-rank approximation matrix and sparse error matrix to form a dictionary for testing.The nuclear coordinate descent method is used to solve the sparse representation matrix.In order to solve the problem of less training samples and the complexity of sparse representations,this algorithm makes full use of the nonlinear information of the original data samples,and maps the training samples into the high-dimensional kernel space through the nonlinear mapping defined by the Gaussian kernel function.In the feature space,test samples are reconstructed using the feature information of the training samples.The experimental results on the AR and FERET face databases show that the proposed algorithm improves the recognition efficiency and guarantees the recognition accuracy under the condition of insufficient training samples.