综合利用人脸特征和活动轮廓技术的人脸检测及跟踪算法

把基于静态图像分割的人脸检测技术同活动轮廓技术相结合，提出了一种在视频序列中检测和跟踪人脸的算法．它在Y、U、V联合梯度的基础土实施分水岭运算，并利用人脸的宽高比、颜色、结构等特征进行人脸逐级判别，从而确定人脸住置，然后用基于骨架的活动轮廓进行精确定位和帧间跟踪，从而较好地实现了视频序列中的人脸检测与跟踪问题。算法具有较强的鲁棒性，对于运动比较平缓、脸部轮廓比较清晰的情况，可以进行长时间跟踪。本算法在面向内容的可伸缩视频压缩编码系统中得到应用。

A Face Detection and Tracking Algorithm Combining Facial Character with Active Contour

Combining face detection technique, which is normally used in static images, and active contour, an algorithm was proposed to locate and track human face in video sequences. The watershed transform is firstly performed on the Y-U-V co-operated gradient, then a series of filtration is carried out to locate the face, including ratio of width to height of candidate face region, color distribution and face structure. Finally, an active contour is used to segment the face from background accurately, and to track the face through frames.