A Video-based Real-time Face Recognition System

Face recognition methods have been the hotspot of the research in pattern recognition field all over the world since face recognition by computer can be widely applied on many aspects in the society in the future. Face recognition is the base line in this paper. Following this base line, face detection and location, illumination and position normalization, feature extraction and selection are studied and a real-time video-based face recognition system is designed and applied. In this system, the video images are captured by USB camera at first. Then a boosting classifier with Haar-like feature is trained and used to detect faces from the video. After the processing of illumination and position normalization, facial feature is extracted by Gabor filter and PCA+LDA. At last, the face from the video is recognized by a distance classifier with trained templates using a pre-determined strategy.This thesis firstly introduces the research background and the technology of face recognition briefly. This part includes geometry feature, template matching method, eigen-faces method, facial information of low frequency method, neutral network and support vector machine. Then several face detection methods are presented including the method based on skin color, the method based on template matching method, the method based on eigen-faces and the method based on neutral network. A boosting classifier is designed to detect faces using OpenCV library. This classifier is based on Haar-like feature. To solve the problem of the bad recognition performance in different illumination conditions, several kinds of illumination normalization methods are introduced in detail and a great deal of experiments are done to compare these methods. Meanwhile, to solve the high dimension problem of Gabor feature, we compare several typical algorithms of feature selection by experiments, and PCA+LDA is used as the solution of reducing feature dimension in our system. Finally, a real-time face recognition system is designed and implemented. This system can achieve a whole process of face recognition. The face recognition system designed and implemented in this thesis has broad perspective, which can be used in the area such as video surveillance, safety entrance, and identity verification etc. It can also be used for entertainment such as video game, intelligent toy and so on.