Face Recognition System Based On Local Feature

Biometric technology is more and more used in identity verification, and face is the most effective part of the authentication. It is suitable for active search in public places and no direct aggression, therefore, it is very meaningful to apply the human face in authentication, the experimental research on the technology of face recognition is very valuable. Face recognition access control system still exists many problems to be solved, such as stability and recognition rate shortage, existence safety hidden trouble and so on, this paper does the research on the existence of the technical difficulties.In this paper, we do something about face detection, in-vivo detection, feature extraction and feature matching algorithm, and the four parts of the study and analysis of the experiment are based on the face recognition system.For face detection, we choose the classic face detection algorithm based on Adboost and complete the algorithm experiment which is based on OpenCV. The experimental results are analyzed, and the stability of the algorithm is good. Multiple faces can be detected in the complex background from the photo and video camera. After detecting the face, the face needs to be determined that it whether is the living face, this paper introduces three main directions about in-vivo detection, and in the face of motion direction of research, introduces three methods of blink detection, analyzes and summarizes the advantages and disadvantages of the three methods, and analyzes the result of the frame difference method. Then in order to solve head and background motion which is interfered with the detection results, we refer to use the literature for statistics the number of connectivity in the eye region mentioned in other people’s papers.For the feature extraction, we use two methods named SIFT and SURF and compare their results in experiments, the results show that SIFT can extract abundant information, but for a long time, SURF extraction information is limited in quantity, and extracting time is faster than SIFT, however multi sample recognition SIFT provided information quantity is also rich enough.For feature matching section, In this paper, the parameter selection of the feature matching method based on Euclidean distance is discussed and analyzed, the Yale face database is used as the experimental sample, we select the best parameters of a feature matching effect. According to The error matching and the complexity of the traditional partition for the face symmetry, in this paper, a simple and effective method based on partition weighted feature matching is proposed. By analyzing the feature matching problem, the problem can be transformed into Feature alignment,and Feature alignment also can be transformed into the problem of finding the optimal coverage of the feature points setted on the face image. Particle swarm optimization algorithm finds the optimal solution by iteration, so this paper makes a basic research on the application of particle swarm optimization algorithm to the feature matching: in this paper, the basic process of particle swarm optimization algorithm is analyzed, and the basic process and parameters of particle swarm optimization algorithm are improved for being applied in the feature matching problem. This method is a new way of thinking and a method of feature matching in face recognition research, it also can effectively improve the recognition rate, but it has the problem of slow convergence.