Abstract: In order to improve the location accuracy and enlarge the application scope of integration projection method in eye locating process, this paper proposes an improved integration projection method based on the edge intensity information in polar coordinate system. The face region is determined according to the skin-color feature. Kirsch operators are used to establish the gray image of the edge intensity information. The angles and directions of eyes can be determined by projecting the edge intensity information to all the polar angles. The polar radiuses of eyes can be determined by differential accumulation of the edge intensity information along the angles and directions of eyes. In this way, the polar coordinates of eyes can be got. Simulation results show that the proposed method can improve the location accuracy of eyes for face images of different poses. Especially, the method can get good robust location results for rotated face images.

Key words: eye location, integration projection, edge intensity, rotation

摘要： 为了提高积分投影方法对人眼定位的准确性和适用范围，提出了一种在极坐标系下对边缘强度信息进行积分投影的改进方法。基于肤色特征确定出人脸区域，采用Kirsch算子建立边缘强度信息图像，对不同极角方向进行积分投影，确定出人眼角度方向，对人眼所在角度方向的边缘强度进行微分累加运算确定出人眼的极径，从而实现人眼的极坐标定位。实验结果表明，该方法能够有效地提高各种姿态人脸图像中人眼定位的准确性，尤其对于旋转人脸图像的人眼定位，该算法具有良好的鲁棒性。

关键词: 人眼定位, 积分投影, 边缘强度, 旋转