虹膜识别算法分析与研究

发布时间：2019-03-06 07:28

【摘要】：生物识别技术逐渐受到人们的重视和市场的迫切需求,已经被不断的研究与应用。虹膜识别技术是该技术中的一种,在人体生物识别技术中倍受青睐。因其具有的独特优越的生理结构,比如:唯一性、稳定性、防伪性,使之成为当今社会个人身份鉴别的重要手段,而且在国防、安防、电子商务、金融等诸多领域中具有广泛的应用前景。本文对虹膜识别系统进行了深入研究与分析,提出了一种四向扫描法快速精确定位虹膜内外边缘;双极性坐标归一化虹膜区域和该区域模块划分;改进的基于25个方向的二维Gabor滤波器特征提取;改进的基于Ferns分类器方法进行虹膜特征训练和样本匹配测试。以上本文提出改进算法主要集中在对虹膜图像进行预处理、特征提取和模式匹配三个部分的模块研究中,具体的详细叙述:第一,预处理:虹膜定位是虹膜图像预处理模块的关键环节,本文采用的是基于四向扫描法的虹膜内外边缘快速精确定位方法。首先采用基于改进Canny算子和小波变换所提取的边缘进行边缘图像融合获取一个足够封闭的内边缘,然后进行四向扫描分割出瞳孔,在进行重新确定边缘点,计算内边缘圆心半径,然后根据内边缘圆心半径粗定位外边缘圆心半径,最后根据微分、积分运算获取精确的外边缘信息,从而精确的定位虹膜内外边缘;为获取更纯的虹膜纹理特征信息,在进行特征提取前,需要做的预处理包括图像归一化和图像增强,本文在归一化中采用的是双极坐标归一化,然后进行模块划分,尽量选取更多的、更加丰富的、纯的虹膜纹理信息,用均衡化增强处理。第二,特征提取:特征提取模块采取的是基于5*5方向改进的二维Gabor滤波器的使用。从选取的虹膜区域提取图像特征,再对Gabor滤波器的各项特性重新分析和参数设置,对25个纹理方向的特征提取融合,根据特征点进行计算和特征编码,获得数据库,并给出了处理后的相关效果图。第三,模式匹配:采用的是基于Ferns(蕨)分类器的方法。介绍训练集和测试样本的划分,分类器的原理,分析在虹膜匹配中所存在的优点,并与支持向量机分类器进行了比较,还与常用的汉明距离法进行比较,给出三种方法的ROC性能曲线比较图。为了对系统中本文提出的算法进行验证,我们使用了中科院提供的CASIA和英国bath两个虹膜数据库共四万多张样本进行性能测试获得了良好的效果。本文算法的操作平台是在MATLAB软件平台上实现的。
[Abstract]:Biometric technology has been paid more and more attention by people and urgent demand in the market, and has been studied and applied continuously. Iris recognition is one of the techniques, which is very popular in human body biometrics. Because of its unique and superior physiological structure, such as uniqueness, stability, anti-counterfeiting, it has become an important means of personal identity identification in today's society, but also in national defense, security, electronic commerce, Finance and other fields have a wide range of application prospects. In this paper, the iris recognition system is deeply studied and analyzed, and a four-way scanning method is proposed to locate the inner and outer edges of the iris quickly and accurately, and the iris region and the module of the iris are normalized by bipolar coordinates, and the four-way scanning method is proposed. The improved two-dimensional Gabor filter feature extraction based on 25 directions and the improved Ferns classifier method for iris feature training and sample matching test. In this paper, the improved algorithm is mainly focused on the iris image pre-processing, feature extraction and pattern matching in three parts of the module research, the specific detailed description: first, the image of the iris image pre-processing, feature extraction and pattern matching module research. Pre-processing: iris localization is the key part of iris image preprocessing module. In this paper, a fast and accurate localization method based on four-way scanning method is used to locate the inner and outer edges of iris. Firstly, the edge image fusion based on the improved Canny operator and wavelet transform is used to obtain a sufficiently closed inner edge, then the pupil is segmented by four-way scanning, and the edge points are re-determined. The inner edge center radius is calculated, then the outer edge center radius is coarsely located according to the inner edge center radius. Finally, the accurate outer edge information is obtained according to the differential and integral operation, so as to accurately locate the inner and outer edge of the iris. In order to obtain more pure iris texture feature information, pre-processing includes image normalization and image enhancement before feature extraction. In this paper, bipolar coordinate normalization is used in normalization, and then module partition is carried out. Select more, more rich, pure iris texture information as far as possible, and use equalization enhancement processing. Second, feature extraction: the feature extraction module adopts the use of 2-D Gabor filter based on 5-5 direction improvement. The image features are extracted from the selected iris region, then the characteristics of the Gabor filter are re-analyzed and the parameters are set, the feature extraction and fusion of 25 texture directions are carried out, and the database is obtained by calculating and encoding the features according to the feature points. The relative effect diagram after treatment is given. Third, pattern matching: the method based on Ferns (fern) classifier is adopted. This paper introduces the classification of training set and test sample, the principle of classifier, analyzes the advantages of iris matching, compares it with support vector machine classifier, and compares it with hamming distance method in common use. The comparison diagrams of ROC performance curves of three methods are given. In order to verify the algorithm proposed in this paper, we use two iris databases, CASIA and bath, provided by the Chinese Academy of Sciences, to test the performance of more than 40,000 iris databases. The operation platform of this algorithm is implemented on the MATLAB software platform.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】