基于AdaBoost算法的快速虹膜检测与定位  

High-speed iris detection and localization based on AdaBoost algorithm

作　　者：陈瑞[1] 林喜荣[1] 丁天怀[2] 

机构地区：[1]清华大学深圳研究生院,深圳518055 [2]清华大学精密测试技术及仪器国家重点实验室,北京100084

出　　处：《清华大学学报（自然科学版）》2008年第11期1923-1926,共4页Journal of Tsinghua University(Science and Technology)

基　　金：清华大学"985"基础研究项目

摘　　要：针对目前虹膜检测和定位方法中检测速度和定位精度等方面存在的不足,提出了一种利用A daBoost算法进行虹膜快速检测和定位的方法。根据虹膜灰度图像的空间结构特征,提取出3类能反映这些结构的H aar-like矩形特征,从中挑选对虹膜图像有最好区分性的385个特征构成弱分类器,再组合生成强分类器。使用正负样本图像训练后,由强分类器级联组成了一个23层分类器系统。实验结果表明:该分类器系统的检测速度平均可达66帧/s,正样本的识别率约为96%,满足了虹膜识别系统实时性的要求。与其他方法相比,有更高的检测速度和定位精度。An iris detection and localization method was developed using the AdaBoost algorithm to overcome the limits of iris detection speeds and localization accuracies with the conventional methods. Based on the space structure properties of iris gray images, three types of Haar-like rectangle features were extracted from which 385 rectangle features with best dividing capacities were selected to construct weak classifiers. The weak classifiers were then combined to build strong classifiers that cascaded a classifier system with 23 layers after training using samples of positive and negative images. Experimental results show that the cascaded classifier system has an average detection speed of 66 frames/s and a positive sample recognition rate of 96%, meeting the real-time requirements of the iris recognition system. Compared with other methods, this method has higher detection speeds and localization accuracies.

关 键 词：虹膜检测 虹膜定位 ADABOOST算法 Haar-like矩形特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]