基于结构和统计信息融合的人脸检测

人脸检测为人脸识别的第一步，是该项技术实用化必须解决的问题。本文提出了一种综合人脸结构和灰度统计特性的新的人脸检测算法。首先，在原始灰度图像上计算了各像素点的梯度方向对称性高的点为特征点，通过梯度幅值分割滤去了大部分孤立的非人脸部件的特征点，再运用一组规则对各个特征块进行组合得到候选人脸区域；然后运用隐马尔马可夫模型对这些侯选人脸的奇异值特征进行识别达到人脸检测的目的。实验表明，本文算法的有效性。

Face Detection Based Combination of Structural and Statistical Information

Face detection is the key to a fully automatic face rec ognition system. A novel algorithm for face detection is presented. It combin es the structural and statistical information of a face. Firstly, the gradient orientation symmetry of every pixel is calculated while the gradien t magnitude is considered. Then according to a set of empirical rules, the obtained eye blocks are mutually grouped. Finally, a trained HMM is employed to recognize the singular value features of the fa ce candidates. During the last step, several improvements are made. Experimental results prove that the algorithm is applicable to real face recogni tion system.