复杂背景下基于时间差分的人脸检测算法

本文主要结合序列图像的运动信息和肤色信息进行复杂背景下的人脸检测。首先利用连续三帧间的运动信息进行粗检测，采用自适应阈值法从差分图像中提取出运动区域，再对差分图像进行相与操作，限制搜索范围；然后利用人脸肤色信息，在YCbCr色彩空间内检测出肤色区域；最后利用人脸的几何特征信息对肤色区域作进一步验证，得到精确人脸。该方法综合了帧间和帧内的分析结果，实现简单、高效。实验证明，在复杂背景下对光照和其他噪声有较好的鲁棒性。
关键词：人脸检测；时间差分；运动信息；肤色特征
Abstract: A method of combining motion and skin-color information of successive images to detect faces in complex background is presented. First, the motion information between three continuous frames is used for coarse detection to obtain the difference images, from which the motion areas can be detected by self-adaptive threshold method, and then take and operation to limit the searching range. Second, the skin-color areas are detected in YCbCr color-space. Finally, geometry feature information is used to further validate the skin-color areas, which can obtain the fine face. This method combines the inter-frame and intra-frame processing, which is simple and effective to realize, and is proved to be robust to illumination and other noises in complex background.
Keywords: Face Detection；Time Difference；Motion Information；Skin-color Feature