文章研究了人脸识别中的人眼定位这一重要环节.将经过预处理的人脸图像先通过积分投影变换法进行人脸定位,在定位后的人脸区域内用一定的矩形框搜索,给方框内黑色像素进行计数,当黑色像素个数达到某个比例时,将其标记为人眼位置侯选处.然后将小方框适当扩大,用canny算子算出人眼的边缘轮廓,并在候选位置中进一步确定人眼位置.实验结果表明本文采用的方法简单易行,具有较高的准确率和较强的光照鲁棒性.%This paper studies the localization of human eyes, an important step in face recognition. The input image is first preprocessed using the integral projection transform to get face area in the image. Then we search the eyes using a certain rectangular box in the face area by counting the black pixels. When the number of black pixels achieves a ratio, we mark the location as an eye candidate. Then the small rectangular box is slightly expanded, and the edge of the eye is calculated with the canny operator to further define locations of the eyes. Experiment results show that the method has high accuracy and strong illumination robustness, and is very simple.
展开▼
