一种眼睛特征抽取的新方法  

A Novel Method for Eye Features Extraction

作　　者：郑忠龙[1] 杨杰[1] 戈新良[1] 杜春华[1] 

机构地区：[1]上海交通大学图像处理与模式识别研究所,200030

出　　处：《上海交通大学学报》2005年第12期1967-1970,共4页Journal of Shanghai Jiaotong University

基　　金：上海市科委重大项目资助(03DZ14015)

摘　　要：眼睛特征对诸如人脸识别、表情识别及三维人脸重建等应用领域而言都是非常有效的信息.本文综合利用彩色信息、G abor小波特征信息以及各特征之间的相互位置关系抽取眼睛特征.首先,在HSV彩色空间的H通道中检测瞳孔的中心位置并对眼球的半径加以估计;然后,在G abor特征空间中构造了眼角滤波器用以检测眼角位置;最后,在前面两步检测的基础上,用样条函数拟合眼睛的轮廓.在SJTU库上的大量仿真实验验证了该方法的准确性与稳健性.This paper presented a novel algorithm to extract eye features, including pupil center and radius, eye corners and eyelid contours, from frontal face images. Such features are very useful cues for applications like face recognition, facial expression recognition and 3D face modeling from 2D images. The novel method is based on color information, Gabor features and the mutual localization relationship between different features. It works in three steps： ① Pupil center is detected and estimated in H channel of HSV color space, and then pupil radius is estimated and refined; ② Eye corners are localized using eyecorner filter based on Gabor feature space; ③ Based on the first and second steps, eyelid curves are fitted by spline function. The experimental results on SJTU dataset show sufficient accuracy and robustness of the novel method.

关 键 词：眼睛特征检测 Gabor特征空间 眼角滤波器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]