Research And Application Based On Face Feature Ectraction

With the development of Internet technology and video technology,extracting the effective content from a large amount of information becomes the focus of a research, and the face region is usually regarded as the region of interest in the video sequence. In a face-related system, it is generally necessary to first perform face detection to detect whether there are faces in the video, determine the number and position of faces, and further determine the precise position of the different regions in the face by precise positioning of the feature points. Finally, according to the previously acquired facial feature information, develop the related applications.In view of these problems and the use of scenes (face detection, virtual glasses try based on face features extraction) this paper did the research on the related technology. And this paper propose an improved algorithm based on the basic algorithm to complete a real-time face detection precision Algorithm and the application method of virtual glasses with high robustness. The main innovations of this paper are as follows:1) An improved AdaBoost methods to detect faceThis paper proposes an automatic and robust method to detect human faces from background that is capable of processing frames of video sequence rapidly while achieving high detection rates regardless of scale,rotation and shelter. The field of this work is the incorporation of AdaBoost and self-adaption region of interest to reducing time complexity in testing time. Next template matching is aim to alleviate common problems in conventional face detection using AdaBoost method such as: inconsistent performance due to sensitivity to rotation of head pose and the obscured of other object. In the final step, focusing on the phenomena of overmatching in test time, this paper proposes an error-correction-factor to overcome.2) Determine the position of the glasses using an multi-region feature points modelThis paper designs and implements a virtual glasses try on application based on facial feature point extraction. One of its core algorithms is to determine the wearing position of the glasses by using the multi-region feature points of the face. It is usually easy to associate with the wearing of eyeglasses and the information about the eye. But the author found that the information of eye is not enough to provide high robustness to determine the wearing position, so a model based on multi-region is proposed. In this model, the information of the eye provides information on the region and central coordinates of the wearing position, information on the nasal feature points and information on the feature points of the mouth to determine the deflection angle that the spectacles need to occur.Experiments show that the proposed algorithm can adapt to the scale change of human face and a certain degree of attitude change.3) An improved adaptive threshold based on Otsu Algorithm for Extracting Eyeglasses ImagesThe other core algorithm proposed in this paper is to separate the foreground and background of the glasses image to get the image to be worn. In order to achieve the balance between the effect and the time complexity, this paper uses the mask method, which requires an adaptive algorithm to obtain binarized images of glasses images. Traditional Otsu algorithm can not be very good in the situation that background and the prospects of the pixel ratio is far different, so Otsu algorithm for this specific situation is limited.