Application Of Graph Embedding Dimension Reduction Algorithm Based On Sparse Representation In Face Recognition

In today's society, people see the public security issues very important, and in the field of security biometric identification technology is playing a more and more important role. As a typical representative of biometric technology, face recognition technology has been widely used in the field of monitoring system, information security, identity authentication and identification. Due to the face recognition technology in practical application in the process of face images often suffer effect of uneven illumination and occlusion factors. These factors will reduce the precision and accuracy of face recognition, which directly leads to the serious decline in recognition rate.According to the technical requirements of face recognition process, the use of suitable wavelet transform and data dimension reduction algorithm for face image dimension reduction transform processing, in order to obtain a higher recognition rate of face recognition. This paper mainly carried out the following aspects of the work:(1) the sparse said method and its composition and research based on graph embedding dimension reduction model; the sparse representation of its concept, dictionary construction and composition were studied; based on graph embedding dimension reduction model includes two methods for constructing the graph embedding model and graph, this paper selects the k nearest neighbor method and epsilon ball nearest neighbor two graph method of composition analysis.(2) proposed a method based on sparse representation of relationship between samples in an attempt to form for the establishment of new based on sparse representation of the label propagation algorithm, in order to in the semi supervised learning algorithm has a large number of unlabeled samples and a small amount of labeled samples to solve. Is proposed in this paper based on sparse representation of the label propagation algorithm mainly includes two steps: firstly, the norm minimization method to the sparseness of the sample between said diagram were established; secondly, through sample labeling tag information and the weight matrix s to calculate the unlabeled samples information.(3) were the label transfer algorithm for face recognition algorithm and the traditional comparative analysis based on the Euclidean distance of the k-nearest neighbor method. It is proved that the the label transfer algorithm for face recognition algorithm can effectively solve the face recognition technology suffer effect of occlusion and uneven illumination problems in practical application process, so as to ensure the recognition accuracy, improve the accuracy of recognition algorithm.(4) were presented in this paper based on sparse representation of graph embedding dimension reduction algorithm in face recognition applications, and after various recognition framework in different categories of 10 ORL subset classification experiments proved that the proposed the LPSR+ NMF-L1 overall framework to obtain the optimal identification effect. It can be proved that the proposed algorithm based on sparse representation in this paper has better application effect in face recognition.