Research On Dimensionality Reduction Of Uncontrolled Face Image Based On Sparse Map Mapping

In recent years,with the continuous development of computer technology and the upgrading of image acquisition equipment,and face recognition technology is mostly used in uncontrolled environment,processing of face images affected by expression,posture,hairstyle,accessories,occlusion,age and so on,and it is complex and changeable in high-dimensional space,which leads to some traditional dimensionality reduction algorithms for controllable face recognition that do not perform well in uncontrolled face recognition.In order to solve this problem,this paper studies the dimensionality reduction of the high-dimensional face image obtained in the uncontrolled environment,and expects to find a mapping or projection to obtain the low-dimensional nature structure of the high-dimensional complex data,which is implicit in intra-class compact and interclass separation,to improve the robustness and accuracy of uncontrolled face recognition algorithm.The specific work contents are as follows:(1)In view of the shortcomings of sparsity preserving projection(SPP)in sparsity reconstruction,such as neglecting the class information of training samples and insufficient discrimination,this paper proposes Weighted Discrimination Sparsity Preserving Projection(WDSPP)algorithm,which can enhance the reconstruction relationship between the query sample and the samples of the same kind by adding the sample class label and the intra-class compact term,and the sparse reconstruction coefficient is constrained by using the weight matrix of the sample distance to reduce the influence of the singular samples in similar samples.In the low-dimensional projection,the global constraint factor is added,and the identification information implicit in the global distribution of the sample is used to make the low-dimensional subspace distribution more compact and easier to identify.A large number of experiments on the AR,the Extended Yale B,the LFW and the Pub Fig databases show that the WDSPP algorithm proposed in this paper has better accuracy in face recognition in uncontrolled environment.(2)In the process of dimension reduction,WDSPP algorithm pulls the image into column vector to reduce dimension,which destroys the spatial structure of the original image to a certain extent,and causes problems such as dimensional disasters and small samples.To solve this problem,this paper applies tensor representation to WDSPP algorithm,and proposes Weighted Discriminant Tensor Sparsity Preserving(WDTSP)algorithm.On the one hand,the WDSPP algorithm is used to calculate the reconstruction relationship between the high-dimensional second-order tensor samples based on the reconstruction idea of compact constraints and weighted constraints in intra-class;On the other hand,based on the reconstruction relationship of the nearest neighbor of the high-dimensional second-order tensor samples,the optimal bilateral projection matrices U and V are optimized to obtain a more discriminative low-dimensional second-order tensor quantum space.The experimental results of four face databases,such as AR and LFW database,show that WDTSP algorithm based on tensor representation is more accurate than WDSPP algorithm based on vector representation.(3)The traditional dimensionality reduction method is a global analysis of the entire sample,which is not effective for the face image collected in uncontrolled environment.To solve this problem,Block Weighted Discriminant Tensor Sparsity Preserving(BWDTSP)algorithm is proposed.Firstly,the image is divided into blocks,and the sub-tensor samples are discriminantly analyzed using the WDTSP algorithm,which can more effectively capture the local information of the face.Then,the residual of each block subset is calculated and fused by SRC classifier,which is used for classification and recognition.The experimental results show that,compared with the WDTSP algorithm,the BWDTSP algorithm has better recognition accuracy in face recognition in an uncontrolled environment.