基于眼裂的人脸图像归一化

Palpebral Fissures Based on Human Face Normalization

针对人脸识别中图像前期处理的人脸图像归一化问题,用AdaBoostCascade方法检测出人脸和眼裂,提出以两眼裂中心点连线为基准进行水平旋转和尺度的归一化矫正方法。与通常的利用两眼瞳距方法相比较,该方法更加快速准确,能处理多达20°的倾斜,对戴眼镜、图像模糊、俯视、仰视、斜视等情况的鲁棒性更好。

A novel approach is presented in this paper, using the line between the two center points of the palpebral fissures of two eyes to normalize the face images for the image pretreatment of face recognition. The AdaBoost Cascade method is used to detect faces and palpebral fissures. It could deal with tilt from 0°to 20°, and get more robust outcomes than that using the line between two pupils commonly in dealing with the image being veiled, overlooking,looking up, sidelong and having on glasses.