Research On Collaboration AdaBoost Multi-feature Multi-pose Face Detection

In the field of Pattern Recognition and Computer Vision, face is a very important biological characteristics, related information processing technology, especially the face detection technology has achieved great development. However, there are still many problems for further research. For example: attitude, illumination, occlusion, expression and other factors still heavily influence the effect of face detection. In this paper, some issues, such as the improvement of detecting-speed, multi-pose face detection and others have been deeply studied.The main study contents and innovative work are shown as follows:(1) In this paper we focus on AdaBoost algorithm first, which is a hot algorithm in the field of face detection.Including the use of Harr-like characteristics, integral figure strategy, Classifier cascade strategy and the search strategy of algorithm. In the process of research we found that AdaBoost algorithm need to search all the areas of the whole image, but the face region accounts for only a small portion in an image, which means that we use a long detection time to detect many useless areas, that is to say we waste too much detection time.(2) Aimed at the problem of AdaBoost all-over search an image, which waste detection time seriously, skin detection is proposed in this paper to optimize the front end of AdaBoost face detection. Skin detection have a fast detecting-speed, and with relative stability.And then in segmentation, every pixel's value of Cb and Cr will be tested and classified in YCbCr space, the segmentated pixels make up the skin-color area, which will be detected by AdaBoost next. The experimental results show that this method is 29% faster than tradition AdaBoost face detection, meanwhile, it also show that this method is still exist defects, while detecting multi-pose face especially excessive rotation angle face, many face being missed, which is a very serious problem.(3) In order to solve the problem of multi-pose face detection, in front-end optimization basis, a new algorithm called Collaboration AdaBoost is developed for multi-feature multi-pose face detection. In this method, the extended Harr-like feature and edge-orientation feature are both applied. The expression can excavate more abundant information from the original data than individual feature; The whole pose angle is decomposed coarse-to-fine, and a classifier-pyramid architecture is adopted to detect multi-pose faces efficiently. Experiments show that:The multi-pose face detection method of more features fusion in a small size in this paper proposed, solved multi-pose face detection problem efficiently, and also achieved higher precision than single feature method.