Improvement Of Discriminant Locality Preserving Projection Algorithm And Its Application To Face Recognition

Dimensionality reduction is a hot topic in current research.It is a problem that people must solve to reduce the dimensionality of the data,but also to maintain the effective information of the data itself and reduce the loss of data information.Discriminant Locality Preserving Projection(DLPP)based on manifold learning algorithm is an effective dimensionality reduction method,which can effectively use the discriminant information in the data to find the best discriminant features from the subspace by maximizing the distance between classes.However,it cannot avoid the traditional Small Sample Size(SSS)problem,that is,the dimension of the sample is much larger than the number of samples.In this paper,combined with the Exponential Discriminant Locality Preserving Projection(EDLPP)algorithm,the matrix exponential is used to further propose an improved DLPP algorithm,called the Generalized Exponential Discriminant Locality Preserving Projection(GEDLPP)algorithm.The algorithm constructs a universal matrix exponential function to avoid the generation of singular matrices,thereby solving the problem of SSS,at the same time,when calculating the generalized feature matrix,by orthogonalizing the projection matrix,the original sample data is retained feature information.In addition,the use of GEDLPP implied non-linear mapping makes the distance between samples between classes greatly increased,thereby improving the performance of pattern classification.In order to verify that the GEDLPP algorithm has good performance in pattern classification,this paper selects YALE,Extended YALE-B,CMU-PIE and COIL-20 and other public image databases for experiments.Experimental results show that the GEDLPP algorithm has excellent performance in pattern classification and image recognition compared with other classical dimensionality reduction algorithms,and this algorithm address the problem of SSS and has better robustness.In addition,this paper applies the GEDLPP algorithm to face recognition and implements it in software.Based on the MATLAB software platform,the necessary functions in the face recognition process are realized.The software can be applied to smaller-scale scenes,such as face recognition and recognition with a small amount of data,such as company attendance check and laboratory access control.