Study On Face Detection Algorithm Of Image And Video Sequence

With the abroad use of video surveillance in people's life, face detection is considered more and more important. Face detection algorithm has been developed many years, although scholars at home and abroad have put forward many methods, but some difficult problems are still not been solved well. For example, some difficulty caused by the abrupt illumination, face rotation and so on. Otherwise, detecting speed is a difficulty which is hard to breakthough. Aimed at these problems, this paper mainly researches face detection of image of complicated background and video surveillance.Based on the face detection in complicated background image,this paper researches gaussian model in YCbCr space, and puts forwards of using gaussian model to segment skin color in YCbCr to deal with the abrupt illumination. The detecting image is compressed step by step. To up detecting speed, this paper proposes first using PCA to reduce dimension of the areas after segmentation and than using SVM to classify. Test shows method presented here has good results in accuracy and speed.Based on the above work ,this paper reseaches face detecting algorithm in video surveillance. Analysis the time consuming of AdaBoost algorithm,and puts forward an improved method. To solve a simple classifier h j of a feature f j, its thresholdθj and bias p j must be confirmed. Bias could be discussed in two conditions, it is +1 and -1. This method could up the detecting speed.On constructing feature and sample,this paper presents a method of reducing the number of feature and choosing training sample effective. At last, the paper designs detecting system. Test shows that it has good result.