Design And Research Of Face Recognition And Homomorphic Encryption Scheme Based On Image Subspace And Kernel Sparse Representation

At present,the face recognition algorithm mainly focuses on two areas of face feature extraction and classification and decision making.The popularity of face image subspaces and sparse representation face feature extraction methods has continued to increase in recent years.At the same time,these methods are mostly based on the partial linear feature extraction of face images,and the requirements of unconstrained environments are more complex.The face structure itself is non-linear,and the user's requirements for the system's experience are also getting higher and higher.This requires finding an algorithm with better effect under the condition of non-linearity.At the same time,with the arrival of big data and cloud computing era,more and more user information is completely exposed on all major Internet media.This poses great hidden dangers and security problems,which is also a concern for users.Based on the above description,this paper summarizes two problems:a.How to find a facial feature extraction algorithm with better performance under the condition of nonlinearity.b.How to find a hidden and safe face recognition algorithm after a is implemented.Based on the above described problems a and b,the work done in this paper is as follows:1.In the Maximizing Margin and Discriminant Locality Preserving Projections(MMDLPP)algorithm,it can maintain similar neighbor relations,distinguish different pseudo neighbor relationships,and can also solve the small sample problem.Supervise learning methods.However,MMDLPP is a linear feature extraction method.There will be a mapping of low-dimensional linear inseparable patterns into high-dimensional space to achieve linear separability,and then cause "dimensional disaster" in high-dimensional space calculations.At the same time,the face structure is usually non-linear.How to find a non-linear feature that can obtain the face,but also has the advantage of MMDLPP algorithm becomes a problem.In order to solve this problem,this paper studies the kernel function.The working principle and eigenfaces convert the high-order matrix calculations into low-order matrix computations.Combined with the merits of the MMDLPP algorithm,a nuclearized MMDLPP(KMMDLPP)algorithm is proposed,which projects the original samples into the high-dimensional space through nonlinear mapping.,and then convert the calculation of the high-dimensional space into an eigenvalue decomposition problem.This method not only maintains the intra-class neighbor relations better,but also separates the pseudo-near-neighbor relationships between classes.It also solves small sample problems better.It also has good robustness in unconstrained environments.The work laid the foundation.2.In the sparsity-preserving embedding(DSPE)algorithm based on sparse representation of over-complete signals combined with discriminative information,the algorithm combines the sparse representation of over-complete signals with subspace methods,which is very good Solve the relationship between real and pseudo-neighbors,and also have good robustness in face recognition in unconstrained environments.DSPE is a linear sparse representation method.In order to obtain a nonlinear sparse representation of face images,this paper presents the idea of using nuclear technology to nuclearize sparse matrices,and at the same time obtains KDSPE after nucleating DSPE as a whole.The KDSPE algorithm projects the original sample into the high-dimensional space through nonlinear mapping,and then converts the calculation of the high-dimensional space into the problem of using the kernel function to solve the inner product.As a non-linear supervised learning method,KDSPE not only obtains the global optimal low-dimensional embedding of the image,but also effectively extracts the non-linear features of the face.At the same time,it has good robustness in an unconstrained environment.3.With the arrival of the era of big data and cloud computing,more and more user information is completely exposed on all major Internet media.This poses great hidden dangers and security issues.In order to protect users' personal privacy information,Completing the secret computing of face image data on the server side,this paper proposes a secret face image based on Kernel Discriminative Sparse Keeping Embedded Algorithm(KDSPE)combined with homomorphic encryption in cryptography and inadvertent transfer protocol based on Identity Encryption System(IBE).Identify the program.The terminal collects the data of the sample to be tested and the face image data of the database to compare,so as to judge whether the face data collected by the terminal exists in the database.Here,the core discriminant sparse matrix obtained by the KDSPE algorithm is used,and then the homonym encryption and the inadvertent transfer protocol based on the identity encryption system(IBE)are used to covertly calculate the Euclidean distance of the core discriminative sparse matrix between the terminal and the server face,thereby judging whether or not there is a match.The advantage of this solution is that it can not only effectively extract the non-linear features of the face,but also has good robustness in the non-constrained environment(attitude,expression,lighting,occlusion,age,shooting angle),in addition to the combination of cryptography.The knowledge,the program can also guarantee the security of the communication participants' data and the security of the communication channel.The experimental results show that the proposed scheme improves the face recognition rate and has certain algorithm security.