Abstract: Considering the proportion of different chroma of skin color，a method of skin detection based on improved YCbCr color space is proposed.Under simple skin underground，changing axis on YUV color space，in order to increase distribution deference between skin samples and lip samples，extracting the center of mass and direction of rotation from lip area，and finally a strategy is proposed based on the knowledge about facial structure/distribution to locate face.Experimental results prove that skin pixel has better clustering in improved YCbCr color space.The method of lip location can provide lower false rate and lower complexity suited to reality.In addition，the method has good detection performance in multi-face image.

Key words: skin detection, face detection, lip extract

摘要： 根据肤色色度的分配比例，提出了一种改进YCbCr的Bayesian肤色检测方法，在肤色背景下，利用肤色和唇色在YUV空间分布特点，变换YUV空间的坐标轴，增大唇色和肤色V分量上的差异。提取了唇色的质心和旋转方向，根据人脸几何特征的先验知识建立人脸定位模型。实验表明在改进的颜色空间内，肤色投影到CrCg平面内得到更好的聚类效果。唇色定位人脸算法简单，速度快，更具实效性，对旋转的人脸同样有效。

关键词: 肤色检测, 人脸检测, 唇色提取