Abstract: Integrating with skin color and facial feature, a fast method was proposed for detecting and locating the face. In normalized RGB color space, skin pixels and lop pixels could be extracted by skin lotus. If a candidate face region did not contain lip pixels, it could be eliminated. The intensity of eye region is higher than that of others in the B component of normalized RGB color space because of sclera and pupil glisten. A region growing method was used to produce potential eyes, and then the geometry feature of eyes and lip were used to locate the face. The results of experiments indicate that the proposed method is robust and effective.

Key words: skin color, face, detection, location

摘要： 结合肤色信息与人脸几何分布特征，提出了一种快速的基于人脸特征的检测与定位方法。在标准RGB彩色空间，通过肤色轨迹进行肤色像素与嘴唇像素提取，根据提取的肤色区域是否有嘴唇像素可初步排除一些非人脸区域；利用人眼较高的蓝色分量及瞳孔反光形成的亮斑，在标准RGB色彩空间的B分量图中通过区域增长法产生潜在眼睛区域，利用人眼与嘴唇的面部几何分布特征，提出一些新的规则判断提取的肤色区域是否为人脸，如果是人脸，则对眼睛进行定位。实验结果表明，提出的方法是健壮的、有效的。

关键词: 肤色, 人脸, 检测, 定位