基于灰度信息和支持向量机的人眼检测方法  

Eye detection method using gray intensity information and support vector machines

作　　者：于明鑫[1] 周远松 王向周[1] 林英姿[3] 王渝[1] 

机构地区：[1]北京理工大学自动化学院,北京100081 [2]北京科技大学土木与环境工程学院,北京100083 [3]美国东北大学工程学院,波士顿02148

出　　处：《工程科学学报》2015年第6期804-811,共8页Chinese Journal of Engineering

基　　金：国家留学基金委资助项目(201306030055)

摘　　要：提出一种基于灰度信息和支持向量机的人眼检测方法.首先,利用人眼区域灰度变化比人脸其他部位灰度变化明显的特征,采用图像灰度二阶矩(方差)建立人眼方差滤波器,在固定人眼搜索区域内,应用人眼方差滤波器搜索候选人眼图像;然后,使用训练的支持向量机分类器精确检测人眼区域位置;最后,采用图像灰度信息率定位人眼中心(虹膜中心).该方法在Bio ID、FERET和IMM人脸数据库中的测试结果显示:没有佩戴眼镜人脸图像正确率分别为98.2%、97.8%和98.9%,406幅佩戴眼镜人脸图像正确率为94.9%;人眼中心定位正确率分别为90.5%、88.3%和96.1%.通过与目前方法比较,该方法获得较好的检测效果.This article introduces an efficient eye detection method based on gray intensity information and support vector machines （SVM）. Firstly, using the evidence that gray intensity variation in the eye region is obvious, an eye variance filter （EVF） was constructed. Within the selected eye search region, the eye variance filter was used to find out eye candidate regions. Secondly, a trained support vector machine classifier was employed to detect the precise eye location among these eye candidate regions. Lastly, the eye center, i. e. , iris center, could be located by the proposed gray intensity information rate. The proposed method was evaluated on the BioID, FERET, and IMM face databases, respectively. The correct rates of eye detection on face images without glasses are 98.2% , 97.8% and 98.9% respectively and that with glasses is 94. 9%. The correct rates of eye center localization are 90. 5% , 88.3% and 96. 1%, respectively. Compared with state-of-the-art methods, the proposed method achieves good detection performance.

关 键 词：模式识别 人眼检测 支持向量机 灰度 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]