复杂背景下人眼定位及人脸检测

人眼定位和人脸检测是人脸识别的重要环节，复杂背景下，人眼定位及人脸检测容易受到光照以及其它物体的影响。在没有检测到人脸的情况下，对原始图像用sobel算子得到灰度边缘图像，进而得到边缘灰度加强图像，估算分割双眼的阈值范围。对边缘灰度加强图像二值化后，结合人脸的几何特征以及双眼的二维相关系数自动确定双眼候选点，利用人脸模板检验双眼及人脸的真实性。大量实验结果表明，该算法在复杂背景下进行人眼定位及人脸检测是有效的。

Eyes Location and Face Detection in Complex Background

Eyes location and face detection are important parts of face recognition, and in complex background, illumination and other objects have influence on eyes location and face detection. When the face is not detected, sobel operator is on an original image to get edge grayscale image and edge grayscale enhanced image, then the scope of threshold that would separate eyes is evaluated according to histogram of edge grayscale enhanced face images. Eyes candidates are automatically selected based on facial geometrical features and 2D correlation coefficient. At last face model is used to check the truth of detected eyes and face. The experimental result shows that the algorithm is effective and robust in complex background.