摘要
针对低分辨率图像中瞳孔边缘模糊、瞳孔受上下眼睑遮挡等问题,提出了一种由粗及精的瞳孔中心精确定位方法,该方法主要使用瞳孔的几何特征和人眼的方向信息进行定位。首先基于显式形状回归方法提取人脸特征点,根据人眼特征点粗定位瞳孔中心;其次根据半径搜索范围,采用射线法提取准确的瞳孔边缘,通过改进的随机抽样一致性算法剔除伪边界点,基于最小二乘法精确定位瞳孔中心。最后本文算法在自制人脸库和公开人脸库BioID上进行了验证,瞳孔中心精确定位准确率为82.2%。结果表明该算法能够克服头部姿势变化和眼睑遮挡等问题,与其他算法相比表现出良好的鲁棒性和准确性。
Aiming at the problem of occlusion by eyelids and the blur edges of pupil in low-resolusion images, a coarse to fine algorithm for accurate pupil center localization is proposed. The proposed method uses the geometrical information of the iris. Firstly, the explicit shape regression method is used to locate the facial landmarks and calculate the rough pupil center. Secondly, the iris edge is detected by starburst with eye direction and the precise pupil center is estimated using circle fitting. The performance of the proposed method is evaluated both in selbmade database and in public database BioID and achieves an accuracy of 82.2~. Experimental results show great improvement compared to the state of the arts.
作者
闫蓓
吴梦瑶
Yan Bei;Wu Mengyao(School of Automation Science and Electrical Engineering,Beihang University,Beijing 1001!）1,China)
出处
《电子测量技术》
2018年第16期74-78,共5页
Electronic Measurement Technology
关键词
瞳孔检测
人脸特征点
射线法
随机抽样一致
圆拟合
pupil detection
facial landmarks
starburst
random sample consensus
circle fitting
