Muti-pose Face Detection Based On Skin-color Segmentation And Adaboost Classifiers

Face detection, as one of the most basic research subjects in face image processing and face research, has an important significance and a broad application in vision and image processing, pattern recognition, video Surveillance and other fields.There has been plenty of research on the face detection arithmetic in recent years. However, most of them, which are carried out, contrapose frontal faces. In this paper, the simple review and analysis on the actual face detection arithmetic are put forward. A multi-pose face detection algorithm is designed based on the skin-color segmentation and AdaBoost face detection algorithm, in which face detection is divided into two procedures—the skin-color region extraction and face detection:First, the input image is pretreated. Then, the skin-color segmentation is carried out for the being tested images based on the established skin model and the segmented skin-color regions are preliminarily screened to obtain face candidate regions. Second, the obtained face candidate regions as the sub-window of the new being tested image are further detected by a group of multi-pose face detection classifiers and the face regions are labeled. The main work accomplished in the paper includes:(l)The method and procedures of skin-color segmentation is introduced in the stage of face region screening via skin-color, including:pretreatment of images, choosing and establishing of the skin-color models, segmentation of the skin-color regions and face region screening. (2)The face detection arithmetic based on AdaBoost and the related concepts are introduced, mainly including rectangle feature, calculation of eigenvalue by integral image, all kinds of face classifiers and training procedures. Meanwhile, on the basis of quondam rectangle feature the new one is contained and the corresponding computing method of the eigenvalue is proposed.(3)Based on the skin-color segmentation and AdaBoost algorithm the implementation process of the overall face detection algorithm is presented, on the basis of which the design of the multi-pose face detectors is introduced:face candidate regions are successively screened and eliminated by multiple cascade classifiers, and the final detection results are accumulated.(4)In addition, a mass of multi-pose face samples and non-face samples are collected in the paper, and the multiple face detection classifiers for different poses are trained by AdaBoost algorithm, respectively.(5)Finally, the experiment scheme of face detection algorithm is designed and related face detection experiments are performed:the face detection algorithm proposed in the paper is compared with that based on the skin-color and that based on AdaBoost, and the statistics and analysis of experimental data is executed. At last, the merits and demerits of the face detection algorithm are summarized and the further work is put forward.