Face Recognition By Improved Elastic Matching

Face recognition is a non-contact, little of violation and easily accepted method in the field of recognition in biometric technology, the research of it has not only the theoretical research value but also wide applying prospect, and it is the meeting-point of applied mathematics, image processing and pattern identification.In this thesis, the current face recognition algorithms and the applications of face recognition are summarized firstly; also the six frequent used methods for face recognition. The face recognition algorithms based on elastic graph matching and the generalized principle component analysis (GPCA) are the research priority. In addition, some improved methods of the elastic graph matching are proposed. The main content of this paper is as follows:First, the elastic graph matching face recognition algorithm based on significant point is introduced, and implemented in ORL database.Second, realize the algorithm based on Gabor wavelet filter. Through two dimensional Gabor wavelets transform, characteristic matrix is shaped by feature extraction, without the decline of recognition rate, wavelet is applied to reduce face matrix dimensions for the simple calculate. Generalized principle component analysis is introduced. An improved method is also given by doing two generalized principle component analysis in the horizontal direction and the vertical direction. The experiments of the two methods carried out in ORL database shows that the recognition rate of the new algorithm is very well.Third, take the advantage of both methods, a new method based on the combination of Gabor wavelet transform and generalized principle component analysis (GPCA) is proposed. In this thesis, the concrete steps are showing. The wavelet is also applied to reduce face matrix dimensions for the simple calculating. The experiments in ORL database shows that the new algorithm improves the recognition rate have superiority over either of the two methods.Forth, the algorithm based on the Gabor wavelet for feature extraction and RBP neural network as classifier, take full advantages of RBP neural network simple structure, muscularity approach of nonlinear phase-matching, global convergence and its high speed of convergence. It is good for faces classifying, while feature extraction is based on Gabor wavelet transform, and the experiment results show that it is a good algorithm.