Integration Of Local Linear Embedding And Linear Discriminant Analysis For Face Recognition Technology Research

For decades, Face Recognition technology has been an active topic in image processing and pattern recognition. Its significance lies on both theoretical research and practical application. Local Linear Embedding provides a new idea to process the nonlinear dimemsion reduction problem, which can provide the close denotation of low dimension coordinates for the high dimension data. Thus it can be used as a tool of dimension reduction when researching the high dimension data and it has a tremendous application potential.This dissertation focuses on face recognition of fusing the nonlinear and linear facial features. The method is proposed according to the disadvantage of linear methods which can not extract the nonlinear structure of data effectively. Firstly, the nonlinear and linear facial features are extracted from the original images individually. Secondly, in order to extract more useful facial feature information, we use the technology of feature fusion to fuse the feature extracted by manifold algorithm LLE and feature extracted by LDA to avoid both their shortcomings. Finally, according to the fusing feature, we select the max matching score as the result of classification.The dissertation proposes the face recognition method fusing LLE and LDA, which realize the feature fusion of nonlinear and linear. For adding the linear and nonlinear features, the research has found that this method can achieve better recognition rate compare to using LLE, LDA individually or combine them simply.