Research On Face Detection Technology Based On AdaBoost Algorithm

Human face detection is a vital part of face recognition technology, and it is also a typical application in object detection technology. As the development of information technology the human face detection becomes more and more important with it extensive utility. It is significant to human face recognition, image searching, vision surveillance, facial expression analysis etc. In this paper, we focus mainly on the visible video image to carry out our face detection method. Considering the utility, real time property, meantime we learn the AdaBoost human face detection algorithm in details, and bring out a new face detection system with relatively a high utility and satisfied real time property.The experimental result latter show the proposed algorithm is satisfactory. We use the AdaBoost learn training algorithm mainly in this paper, by training the classifier with right and wrong samples, to make it perform the right face detection function. As for a face detection system which need lots of data to train it in order to get accurate parameters, the training time cost is non-neglect. The efficiency of a face detection algorithm determined the overall time cost of program debugging. Therefore, in this paper, aimed at the relatively high time cost in the training phrase of the AdaBoost learn algorithm, we make some improvements, and bring out a new method, which calculate the error rate directly, avoid the time consuming iterated arithmetic and probability distribution, statistical calculation etc. As a result, the training speed and overall efficiency of the algorithm is improved by large.In order to improve the detection efficiency of the overall system, by considering the traditional face detection method's highly calculation complexity in the scanning period, we added a waterfall framework, using a multi-layer structure to eliminate the non-facial parts by layers, therefore, lessened the detection area, and also, avoid unnecessary calculation as far as possible. Thus, make the face detection speed and efficiency achieved a great improvement, thereby, enhanced the utility and real time property of this system.To achieve relatively a high face detection robustness, and extent the utility field, we added a static face detection, which rotation angle is no more than 40 degree, and its time request is not high relatively. Considering the significant of human eyes'position, at the same time, we added a human eyes detection modular, thus, make preparation to the following face detection process.The experimental computer property is: CPU: core due, 1G RAM. The detection speed is 14 frame per second in the video detection, which image resolution is 320×220 pixels, and the input face pose rotation is limited within 15 degree. If the input is static image, the detection algorithm could process an image with higher pixels than the video images, also the static face pose rotation could limit within 30 degree.