Research On Face Detection Method And Its System Implementation

Face detection is that, for any given image, using some of its searching strategies to determine whether it contains any faces, if it is then track face location and size. As the basis of face recognition and face expression recognition, face detection is an important application of pattern recognition. It has profound application prospects in security recognition and identification. Initially, the topic of face detection is raised on the needs of face recognition. But with the application of further development, it has become a independent subject. Now, face detection has great value in authentication and visual monitoring area.This paper researched on how to detect human faces in images and videos quickly and accurately. It built a prototype system of face detection based on Adaboost algorithm, which consisted of two independent parts: training samples to get a classifier part and face detection part.What had been done in this paper is as follows:1. Trained samples to get a strong classifier based on Adaboost algorithm. Haar-like rectangle features were been used in the training process and then cascade those strong classifiers to a multi-layered one. To improve the system robustness, in train process, the rectangle features with 45 degree tilt were used, and the number of training samples was increased by adding samples of faces in different angles and samples of faces with attachments. Set selecting mark for every sample to avoid some samples being selected repeatedly.2. The method of zooming out the windows is used in detection to reduce computation. Meanwhile, Canny edge detection is used to remove part of the background region, which significantly speed up system detection process.3. When detect human face in video, the skin color detection method is used to reduce the number of sub-windows which will be sent to the multi-layer classifier. So the detection time is shortened, which can satisfy the consuming of time in real -time detection. Besides, the detection rate is increased and the false detecting rate is decreased. For profile face detection in videos, it is combined with accurate detection based on cut-set algorithm in graph theory.