Face Recognition Research Based On Linear Projection Analysis And Kernel Function Methods

Face recognition technology is a very active subject in the field of pattern recognition, which has a wide range of potential applications, such as commercial, law enforcement, public security, identity classification, entrance control, video surveillance and so on. The primary task at hand, given still or video images, requires the identification of one or more persons using a database of stored face images.The research background, significances, present status, and development are denoted in this thesis. The proper preprocessing of face images has been discussed. Linear and nonlinear kernel projection algorithms applied for feature extraction are focused on. With using regularization parameter, KDDA (Kernel Direct Discriminant Analysis) shows greater cluster capability than traditional methods applied on ORL face database. Small Sample Size problem and classifiers are systematically researched as well, and then experiments about "SSS" are made on excessive occasion.As for linear projection, Principle Component Analysis and Linear Discriminant Analysis are expounded. Where PCA seeks directions that are efficient for representation, LDAseeks directions that are good at discriminating samples. Experiments on ORL face database use the two algorithms above to extract features. Results have been discussed in detail.Furthermore, kernel methods are introduced to extract features of face images better, including two nonlinear projection methods KPCA (Kernel Principle Component Analysis) & KDDA (Kernel Direct Discriminant Analysis). In the similar way as before, experiments on ORL demonstrate the effective discriminant abilities of KPCA and KDDA.In addition, the theories of various classifiers are proposed in this thesis. Different classifiers are chosen to distinguish samples using extracted features which are obtained by different projection methods. The recognition performances are compared according to certain criteria.At the end of thesis, Small Sample Size problem, as an eye-catching problem, has been discussed. As for this problem, the regularization parameter is introduced into KDDA, and the results of experiments are analyzed.