视频序列中运动人脸的检测与特征定位

针对视频序列中运动人脸的检测与特征定位问题，给出了一种算法，即通过计算并比较人脸区域而不是整幅视频帧的峰值信噪比（PSNR），先从视频序列中找到人脸区域清晰度高且尽可能大的视频帧，然后再进行人脸的准确检测与特征定位。首先利用Adaboost方法检测出人脸的大致范围，再根据肤色模型确定人脸的具体位置并从图像中提取出人脸部分，然后利用基于帧间差的人脸区域的PSNR判断图像清晰度，从而定位出人脸区域清晰度高且尽可能大的视频帧，最后进行人脸检测和特征定位。对实际视频序列的计算表明，给出的算法在时间效率上较高，人脸的检测与特征定位效果较好。

Face Detection and Feature Location of Moving Men in Video

An algorithm for face detection and feature location of moving men in video is proposed in this paper. By calculating and comparing the Peak Signal-to-Noise Ratio(PSNR) of the face region, other than of the whole frame, the video frame which owns high measure of clarity of face region and owns a face region as large as possible is found and then the exact face is detected and its features are located. In the algorithm, an approximate face region is found first by using the face detection algorithm of Adaboost, and then, by using of the skin color model, a specific face region is determined and the face part is cut from the frame; after that, the video frame which owns high measure of clarity of face region and owns a face region as large as possible is selected by judging the clarity from PSNR, which is based on the difference between neighbor frames of face region; finally, face detection and feature location are processed. Our experiment results show that the algorithm we proposed has a satisfied performance regarding calculating time and has good effect for face detection and feature location.