The Research Of Face Recognition Application Based On LPP

In recent years, face recognition technology has been highly concerned about people. Due to the efforts of many scholars, the development of the technology is more mature. After50years of research, it has been made many achievements in the field of Face Recognition. However, factors such as cosmetic facial feature extraction, face recognition has led to many problems still exist. Small sample size problem which is one of the core issues of face recognition, and it is much pay great attention in manifold pattern recognition. In this paper, we are conducting in-depth research, and focusing on the small sample size problem and the complexity of the problem of dense matrices in face recognition. Research work and innovations are as follows:1. Aim to the LPP algorithm can not make use of the discrimination information and other shortcomings, the Discriminant Locality Preserving Projection (DLPP) algorithm was proposed. It has been successfully used as a dimensionality reduction technique to many classification problems. However, in order to avoid small sample size problem in the calculation process, DLPP needs to reduce dimensions, which will lose some important discriminative information. Direct Linear Discriminant Analysis (DLDA) can solve the problem by different diagonalization. Inspired by DLDA, we incorporated DLDA into LPP and propose a novel method of improvement algorithm. Duo to DLDA algorithm is retained valid determination information, the modified LPP algorithm achieves is better than DLPP and LPP in face recognition.2. According to the above description, in order to improve the recognition rate and avoid the problem of the small sample size, we proposed DLPP algorithm. However, the computation of this algorithm involves dense matrices eigen-decomposition which is needs to be addressed. Because the algorithm consumes a lot of time and memory, so that in the data set is large, the algorithm simply can not be applied. Spectral Regression Discriminant Analysis (SRDA) algorithm can save time and memory effectively, meanwhile spectrum regression analysis algorithm using the spectral method can improve the recognition rate. Inspired by SRDA, we proposed a novel improvement algorithm for Locality Preserving Projection:called Spectral Regression Discriminant Locality Preserving Projection (SRDLPP). The experiments show that the method has high recognition rate and low memory consumption in the face database.3. In the MATLAB experimental platform, we selected ORL, YALE, Extend YALE_B as a test face database. Then we contrast the LPP, DLPP, an improved LPP algorithm and SRDLPP algorithm. The experimental has excellent recognition rate.