一种由粗到精的远距离虹膜定位算法

A coarse to fine locating algorithm of long-distance iris

目前常见的虹膜识别算法均是在近距离且需用户积极配合的情况下进行的。远距离虹膜识别技术具有无需拍摄者刻意配合,非侵犯,易使用等优势,属于生物特征识别中的研究热点问题,其核心是在大尺度图像中快速定位出人体虹膜。针对该问题,提出了一种由粗到精的远距离虹膜图像的定位算法,对人脸图像进行采样,利用类Haar特征与Adaboost构建人眼检测器,并结合先验几何信息对得到的候选区域进行筛选,检测出人眼,将人眼位置变换回原始大尺度图像,分割出人眼区域。最后利用前期积累实现虹膜的准确定位。该算法在CASIA v4.0远距离虹膜图像库中可达到91.66%的准确率,具有较好的应用价值。

The commonly used iris recognition methods can only deal with the iris images caPtured in short range and require users to cooperate actively. The iris recognition technology over a long distance has advantages of not requiring users＇ deliberately cooperating, being not aggressive and easy to use. Therefore, it is becoming a hot issue in biometrics. The most important problem in long distance recognition system is how to perform efficiently and accurately iris locating for large scale face images. To solve this problem, this paper proposes a coarse-to-fine and hierarchical iris locating algorithm for long-distance captured iris images. First, Haar-like features and Adaboost are adopted to construct an eye detector. After that, geometrical information is combined to filter out eyes from candidate regions found by the eye detector. Then, the eye location information is mapped back into the original face images to segment out the eye regions. Finally, accurately iris locating is completed by using our previous short distance locating algorithm. In the CASIA v4.0 long-distance iris data- base, the accuracy of iris location can reach 91.66% , so the proposed algorithm has a good applica- tion value.