一种快速眼睛定位与人脸平面旋转校正方法

An Approach to Fast Eye Location and Face Plane Rotation Correction

针对人脸图像平面旋转导致识别效果不佳的问题,提出一种有效的眼睛定位与人脸平面旋转校正方法.首先基于AdaBoost算法训练得到的眼睛分类器从人脸图像中快速确定双眼候选区域,然后根据双眼在候选区中所处位置和所占比例,将候选区域分为宽度相同的左眼子区、中间子区和右眼子区3个子区域,再对包含眼睛的左眼子区和右眼子区分别求积分投影来准确定位双眼位置;准确定位双眼位置后,给出了以图像中心为旋转基准点时旋转角度的计算方法,并结合图像旋转公式实现了人脸图像的平面旋转校正.实验结果表明,该方法不仅能防止人脸倾斜时出现伪特征点,也能避免眼镜内边框和眼镜支架对定位眼睛的影响,提高眼睛定位精度,而且能实现人脸的平面旋转校正,具有非常好的实时性和实用价值.

In order to overcome the problem of bad effect of recognition due to the face plane rotation, an effective method of eye location and face plane rotation correction is proposed in this paper. First the eye classifier based on the AdaBoost algorithm is used to quickly detect the eye candidate region. Then according to the position and proportion of eyes in the candidate region, the candidate region is divided into three sub-regions with the same width, which are the left-eye sub-region, the intermediate sub-region and right-eye sub-region respectively. We calculate the integral projection for the left-eye sub-region and right-eye sub-region to accurately locate the eyes position. After eyes location, the calculation method of the rotation angle is proposed when the center of the image is used as the reference point, and the plane rotation correction of the face image is realized combined with image rotation formula. The experimental results show that the proposed method can prevent the emergence of pseudo feature points, and avoid the influences of the spectacle border and spectacle frame. The method improves the eyes location accuracy and realizes the face plane rotation correction; it has better real-time performance and practical value.