【摘要】 对人脸识别中的关键一步——眼睛定位进行深入研究。提出了一种结合人脸特征间的几何关系和改进的积分投影的方法,实现了快速有效的眼睛定位。首先,对输入图像进行光照补偿、最大类间方差阈值分割法的二值化处理和数学形态学的开运算处理;其次,框选人脸区域;最后,根据人脸特征间的几何关系先缩小眼睛区域的搜索范围,并分别框选出左右人眼区域,再利用改进的积分投影及其统计数据的峰值位置分别确定左右眼球的中心位置,然后输出眼睛定位的结果。实验表明,该方法具有很高的精度,对不同角度和不同方向的倾斜人脸、脸部表情及姿势变化等具有很好的鲁棒性。更多还原

【Abstract】 This paper introduces a Eye-detection way, the key step for face recognition. A method of combining facial geometric characteristics and improved integral projection was presented to implement an expeditious and effective way to detect eyes. Firstly, light compensation and binary of the input image by using the method of Ostu and Opening was made. Secondly, the face region was selected. Thirdly, eyes′ regions were limited by using the facial geometric characteristics, and the eyes′ positions were detected by using improved integral projection and the location of peak values of the statistic data. Finally, output the accurate positions of the eyes. Experiments showed that this method had high accuracy and good robustness for different angles and different directions of faces and facial expression and gestures.更多还原