Face Recognition Based On Sparse Tensor Analysis

A complete face recognition system mainly includes two parts:feature extraction and classifier recognition.Face recognition based on sparse tensor has great application value in security,information security and modern finance.It has attracted the attention of researchers from all over the world,and it's also the content of this thesis.The main work done in this thesis is as follows:?1?The typical subspace learning theory,Sparsity Preserving Projection algorithm and its application in face recognition are studied.The Sparse representation optimization algorithm,as well as Sparsity Preserving Projection algorithm based on Sparse representation and Locality Preserving Projection principle are analyzed,and a large number of identification experiments based on open face database are carried out.Experiments also include face recognition experiments with different SNR noise and with different degrees of occlusion.The experimental results show that compared with the Locality Preserving Projection algorithm,the Sparsity Preserving Projection algorithm has the advantages of higher recognition rate and stronger anti-noise performance,especially when it comes to different degrees of occlusion.?2?The Tensor Locality Preserving Projection algorithm based on tensor subspace analysis is well studied,and an improved Tensor Locality Preserving Projection algorithm based on coarse-to-fine tactics is proposed.Firstly,the training samples are selected by a regularized non-negative sparse representation model,and samples of partial classes are selected according to the size of the representation coefficients.Then,the training samples are finely selected again using the distance based on the ₂regularized representation.On the basis of the finely selected samples,face recognition based on the Tensor Locality Preserving Projection algorithm is performed.The experimental results show that the Tensor Locality Preserving Projection algorithm of coarse-to-fine strategy not only considers the geometric and topological characteristics of the data embedded in the tensor space,but also solves the problem of redundant and high-volume calculations of the training samples and obtains a better face recognition effect.?3?A face recognition method based on multiple sparse tensor subspace classifiers is proposed.Bayesian decision theory and Bayesian fusion algorithm are analyzed.On these bases,the face recognition methods of Bayesian fusion with three kinds of classifiers:Sparsity Preserving Projections,Tensor Locality Preserving Projection and Sparse representation are studied.The experiments are verified on the publicly released face database.The experimental results show that the proposed Bayesian fusion algorithm based on multiple sparse tensor subspace classifiers has higher recognition accuracy and better anti-noise and occlusion ability than the pre-fusion single classifier.