一种新型的人脸眼睛定位方法

Novel eye detection algorithm

提出了一种在复杂背景、光照、姿势变化条件下的人脸眼睛定位系统。首先采用Adaboost方法提取人脸,并提出了解决平面旋转和深度旋转的方法。接着,采用数学形态学提取人脸特征并用各种规则去过滤特征点。然后采用SVM眼睛确认方法确认眼睛对。最后采用Camshift和Kalman滤波进行跟踪。基于IFACE数据库的实验结果表明我们的算法具有很高的眼睛定位准确率并对光照、姿势、复杂背景不敏感。

A general and efficient design approach are presented using Adaboost, mathematical morphology, rules, SVM, Camshifl and Kalman to cope with eye detection. In order to get face in an image, Adaboost algorithm is used. An effective paradigm to cope with big view angle and planar rotated face is proposed. Then, face features including eyes and mouth are extracted using mathematical morphology. Rules are used to filter the candidates. Then, SVM is used to verify eye pair. Finally, Camshift is combined with Kalman to track face and eyes. Using this algorithm, eye-pair is located exactly from image. Experiments conducted on the IFACE database show that the approach achieves excellent performance in detectin rate and are insensitive to illumination, background and white-frame eyeglasses.