Research On Technology Of Human Face Detecting And Tracking

With the development of information technology and the need of intelligent human-computer communication, research on human face problems becomes an important branch of computer domain, and it has much more use values and research achievements. Human face problems consist of four parts: human face detection, human face tracking, identification of human face and the interrelated pose and expression. This paper does research mainly on human detection and human tracking.We analyze and summarize the correlative representative algorithms about human face detection both home and abroad, and present a new method for detecting human face based on profile image, which is a statistics method. Firstly, we extract the profile features of human face and non- human face from training images (including human training images and non-human training images). Then, put them into Support Vector Machines (SVM) to train and classify those features, then use the training results to detect human faces. The arithmetic extracts the profile features,and segments the profiles from both vertical and horizontal directions of the human face's gray image which is based on physiological features of human faces. Then, the profile image is divided into some blocks, and we use the profile features of all blocks as the profile feature of the whole image. In addition, we use the method of lever collection to achieve accurate location of human face.In the field of human face tracking, we contrast several main similarity methods nowadays, and adopt a method of human face tracking based on feature space model. The method establishes the feature space model according to the conversion relations of color images and gray images. Then, we compute the similarity function, and then choose the feature space which can best distinguish the tracked objects from the background. In the feature space, we compute the gray even of the blocks to establish relativity function which matches the images to track.Finally, a lots of experiments are done on static images and video sequences. The experimental results show that in despite of its a bit low speed, the method is robust and accurate, even when the color of the background and the tracked objects are very similar, it can accurately track and locate the object human face. Besides, it has high computing efficiency, and it can offset the disadvantage of human face detection arithmetic.