中文摘要：

针对光照和表情对眼睛定位产生的影响,给出一种基于多尺度自商图和形态滤波的眼睛定位算法。首先利用一种多尺度自商图方法消除人脸图像的光照影响;然后综合运用形态学操作的开闭运算提升眼睛区域特征并采用相关系数法获得眼睛粗定位,对开闭眼不敏感;最后对粗定位区域进行方差投影并二值化后获得精确的眼睛中心点。在JAFFE数据库、YaleB数据库和AR数据库上测试表明本文算法能很好地定位眼睛中心位置,且对光照、表情鲁棒。

英文摘要：

To eliminate the influence of variant illumination and expression on eye location,an eye location method based on multi-scale self-quotient image and morphological filtering is presented. Firstly a multi- scale self-quotient image method is utilized to eliminate the influence of variant illumination. Then the open and close operation of mathematical morphology is integrated to enhance eye features, the coarse location of eyes is obtained by using correlation coefficient, which is insensitive to the status of eye. Finally the center of eye can be precisely given by variance projection and binaryzation. Experimental results on JAFFE database, Yale B database and AR database indicate the proposed method can precisely locate the eye center,and it is robust to the variant illumination and expression.