视频序列中的人脸检测算法

根据视频序列的特性,首先采用背景差分法提取视频中的运动区域,之后对检测的运动区域采用对数扩展的方式调整光照的影响。采用基于YCbCr颜色空间的肤色模型进行检测得到人脸候选区域,并采用一定的阈值去除小的非人脸的区域,再通过在人脸候选区域检测眼睛位置,利用人脸几何特征精确定位人脸。实验表明,该方法提高了检测速度,降低了误检率,可以应用在视频监控等实时系统中。

Research on Face Detection in Video

The fast detection and exact orientation of face region is the important section of face recognition in video surveillance system.Due to the characteristic of video,the movement regions are obtained using background subtraction method.Then logarithmic expansion method is used to adjust the influence of light.The skin-color method based on YCbCr color space is applied to obtain the likely face region and using a certain threshold to remove small non-face region.Finally the face regions are located by eye position.The experiments show that this method enhanced the detection rate that may be applied into the real-time system.