Technology Of Face Recognition Based On Video Surveillance

Face recognition based on video surveillance is a challenging task. It blends many subjects into a unity, such as pattern recognition, video image processing, computer vision, etc. It has a wide range of application environments and broad prospects, and has become a hot topic, because it provides a highly reliable and convenient identification recognition way.Face detection is the premise of face recognition. The symmetric difference method is used to detect the motion information of the frame in this dissertation. Then, an improved rule-based projection algorithm is used to detect faces based on clustering of facial complexion in YCrCb color space. Not only single face can be detected, but also more human faces in a complex background can be correctly detected. Based on the former several frames, the positions of faces in the later frames can be forecasted and tracked because of the continuity and regularity of the motion in video. Because of human motion, facial regions may change. In this paper, adjust genes are proposed to dynamically adjust the regions. In this way, the disadvantages that traditional prediction algorithms can't solve are avoided. Results show that this algorithm of face detection and forecast-tracking not only is not limited by facial expressions, but also partly eliminates the influence of illumination.The faces detected are pre-processed for extracting the features of faces. Firstly, eyes are located in faces and used as the basis of the pre-processing for extracting the facial features. Here, the pre-processing includes illumination regulation, slope correction and normalization of size. Feature extraction can be divided into geometrical feature extraction and algebraic feature extraction. The geometrical features that combine facial global features with local ones are chosen in this dissertation. And they are composed into a one-dimensional feature vector. Not only facial characteristics but also the geometric shape and the location of each other of facial features are considered. Furthermore, only half facial features are chosen so that the amount of calculation is decreased. The processing becomes more convenience.In this dissertation, the weighted similarity function is chosen as the identification method. And the similarity function is improved. Feature variables are given different weights according to the effects of different facial expressions. In this way, both the invariance and the variability are considered. The experimental results show that the problem influenced by facial expression is better solved. The improved method is effective, and has more adaptability of facial slope, illumination and expression change.