Research Of Face Recognition Technique Based On SVM

Face recognition has many virues comparing with other human body character recognition. For example, it is a friendly manner for user, and can not make user feel that it is a burden because the method of sampling is very easy. At the same time, the requirement for modern image communication is getting urgent, so the research of face recognition has a very important signification.In this paper, the contents, difficulties, process and development of face recognition have been firstly summarized. Then the author study following three problems: face-detection, face-representation and face-distinguishing which are belong to face-recognition problem. Based on the research of representative algorithms, the method based on complexion-geometry model has been adopted in aspect of face detection; the method based SVM(Support Vector Machine) has been utilized in face distinguishing; and the method of wavelet has been chosen in face-feature choice.This paper are concentrated in the following aspects: the research on face detection based on eyes-location, the methods of face-representation, the method of SVM-training and the methods of face-recognition based on SVM. The method based on eyes-location is one of the methods based on complexion-geometry model, but it differs from conventional methods: the first is the idea that eyes are the most important features in the five sense organs; the second is the roughness-to-precision strategy which has been used during the process of eyes-location. The method of face-representation based on wavelet has been adopted through analysis of classic methods. A new method to increase the computation speed of High-scale SVM training has been presented in this paper, because the method of SVM has been used in face recognition in this article. The new method is fit for the face-samples which are proved by experiments. Because SVM is to settle two-class problem, but face recognition is a multi-class problem, two methods have been proposed to solve the contradiction, one is the method based on "intra-face space" and "external face space ", the other is the method based on one-class SVM. And theauthor provided results of experimentations and the analysis of these methods.At last, the author summarized the whole work of this paper, and give a future look into the research that might be taken subsequently.