加权K近邻和加权投票相结合的虹膜识别算法  

An Iris Recognition Algorithm Based on Weighted KNN and Weighted Majority Voting

作　　者：孙彩堂[1,2] 张利彪[1,2] 周春光[1,2] 刘小华[1,2] 

机构地区：[1]吉林大学计算机科学与技术学院,吉林长春130012 [2]教育部符号计算与知识工程重点实验室,吉林长春130012

出　　处：《小型微型计算机系统》2010年第9期1846-1849,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60673099;60873146)资助;国家"八六三"高技术研究发展计划项目(2007AA04Z114;2009AA02Z307)资助;吉林省生物识别新技术重点实验室项目(20082209)资助;吉林省科技发展计划项目(20080168)资助

摘　　要：利用彩色虹膜数据库研究了在正常环境下采集到的有噪声的虹膜图像在不同色度空间中各颜色通道的性能,提出了一种决策层融合方法,利用加权K近邻方法和加权投票相结合进行虹膜识别.对于每个单独的颜色通道,根据比对距离找到前K个近邻,利用排序与相似度相结合的方法为这K个近邻赋权值(RSWKNN),作为各通道的决策输出.在综合各通道的决策时,利用各通道的性能,对各通道的决策输出进行加权投票,本文采用了四种基于通道性能的各通道加权方法.实验分析了各种算法的性能,结果表明:本方法能够有效提高虹膜识别的准确率和稳定性.This paper studied the performance of iris recognition based on the color channels from different color spaces using the color image iris dataset,which is characterized by the fact that many of the images were captured under real conditions so as to incorporate some kinds of noise purposely.Then it proposes a decision level data fusion method for iris recognition,which combines Weighted K-Nearest Neighbor and Weighted Majority Voting method.For each channel,the system finds the first K nearest neighbors according to the distance,and then set weights for them employing an algorithm namely RSWKNN as the output of the channel.After that,the weights are weighted summed according to the performance of each channel.This paper proposed an algorithm to calculate the weights of the K neighbors,and four methods for the weights of each channel.The experiments show that,this method can improve the performance of iris recognition effectively.

关 键 词：生物特征识别 虹膜识别 加权K近邻 加权投票 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]