自适应优化Log-Gabor滤波器与动态径向基函数神经网络的虹膜识别  

Iris Recognition Based on Adaptive Optimization Log-Gabor Filter and Dynamic RBF Neural Network

作　　者：刘帅 刘元宁[1,3] 庄述鑫 侯铭楷 陈静 张水涵 LIU Shuai;LIU Yuanning;ZHUANG Shuxin;HOU Mingkai;CHEN Jing;ZHANG Shuihan(Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education,Jilin University,Changchun 130012,China;College of Software,Jilin University,Changchun 130012,China;College of Computer Science and)

机构地区：[1]吉林大学符号计算与知识工程教育部重点实验室,长春130012 [2]吉林大学软件学院,长春130012 [3]吉林大学计算机科学与技术学院,长春130012

出　　处：《吉林大学学报（理学版）》2019年第2期331-338,共8页Journal of Jilin University:Science Edition

基　　金：国家自然科学基金(批准号:61471181);吉林省自然科学基金(批准号:20140101194JC;20150101056JC)

摘　　要：首先,采用Log-Gabor滤波器提取虹膜幅度特征,根据虹膜库的种类,通过改进的遗传粒子群优化算法优化滤波器参数;其次,利用主成分分析法降低维数,进而减少噪声和冗余;再次,构建动态径向基函数神经网络,并通过虹膜幅度特征间的欧氏距离进行虹膜识别;最后,采用多种小型虹膜库与其他虹膜识别算法进行对比实验,实验结果表明,该算法在一对一虹膜识别中正确率更高,ROC曲线更贴近坐标轴,滤波器通用性更好,提高了小型虹膜库的识别率,解决了传统算法学习收敛速度慢、结构通用性差的问题.Firstly,Log-Gabor filter was used to extract the iris amplitude features,according to the type of iris library the filter parameters were optimized by the improved genetic particle swarm optimization algorithm.Secondly,principal component analysis was used to reduce dimensions,thereby reducing noise and redundancy.Thirdly,a dynamic radial basis function(RBF)neural network was constructed and iris recognition was performed by the Euclidean distance between iris amplitude features.Finally,we compared a variety of small iris libraries with other iris recognition algorithms.The experimental results show that the algorithm has higher accuracy in one-to-one iris recognition,the ROC curve is closer to the coordinate axis,and filter is more versatile,which improves the recognition rate of the small iris library.It solves the problem of slow learning convergence speed and poor structural universality of traditional algorithms.

关 键 词：虹膜识别 LOG-GABOR滤波器 遗传粒子群优化算法 动态径向基函数神经网络 欧氏距离 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]