基于视频图像的人脸定位方法

Method of face detection in video images

给出了一种视频图像中的人脸定位方法,该方法首先通过一种平均帧差的方法检测人体运动区域;然后根据人脸分布的特点,采用水平投影的方法在运动区域里分离人脸和身体;最后根据人脸器官的分布规律,利用k-means的聚类算法对人眼进行定位.实验表明,该方法简单、高效,而且不依赖于人脸肤色,在一定范围内能很好地适应光照变化,能满足实时系统的要求.

It has given a human face detection method in video images.This method detects movement region by average frame difference in the first;In the second,according to the rules of human body distributions,we separate the face and the body by a method of horizontal projection in movement region;Finally,basing on face characteristic,we use k-means cluster algorithm to locate eyes.The experiments indicate that our method is simple,highly effective.Moreover our method does not rely on face skin color and are good adaption in Illumination changes at certain range.Of course,our method are also satisified with requests of real-time systems.