虹膜识别特征提取及鉴别

摘要

随着信息社会的快速发展，对安全的需求也日益增长。虹膜识别技术作为一种身份识别，以其很高的可靠性得到人们的重视。虹膜识别系统核心一般由图像采集、虹膜定位、归一化、特征提取及编码和训练识别五部分构成。本文介绍了目前虹膜识别的现状，简单阐述了一些经典的虹膜识别算法和技术，并完成识别系统。

在虹膜的定位阶段，首先对图像进行缩放，在不影响后续处理的情况下减小了处理的数据量，然后采用梯度加权的Canny算法进行边缘检测，再对边缘图像，采用圆Hough定位方法，分别定位了虹膜的外边界。接着采用Radon变换检测直线的方法分割上下眼睑，阈值法除去睫毛干扰。同时也研究了一些文献中分割眼睑和睫毛的方法。

归一化阶段，采用了文献中普遍使用的“Rubber-Sheet”模型，将虹膜归一化为64512

大小的矩形，以利于特征比对。

在虹膜的特征提取及编码阶段，基于信号处理中的空间/频域技术，采用一维Log Gabor滤波器提取虹膜的纹理信息，对滤波结果的实部和虚部分别进行相位量化和编码，同时也对噪声进行处理，获得相应的掩码。

训练识别阶段，采用海明距离度量虹膜之间的相似度，选取最小距离分类器和具有最小错误率的分类阈值形成组合的分类决策规则。整个识别系统主要在中科院V3.0虹膜数据库上进行了测试。

关键词：虹膜识别；虹膜定位；圆Hough变换；Log Gabor小波

ABSTRACT

With the rapid growth of information technology, the demands of information security are ever-growing. As the technology of identification, iris recognition, for its high reliability, gets great attention. Iris recognition system consists of image capturing, iris location, iris normalization, feature extraction and coding and decision training. In this dissertation, the situation of iris recognition is presented. Some practical algorithms and technique are briefly introduced. A system of iris recognition is fulfilled.

In iris location stage, image is zoomed to reduce the data volume with less influence on post processing. Afterword, the algorithm of Canny edge detection, with weighted gradient, is adopted. In the edge image, Circular Hough transform is applied to locate the inside and out boundary of iris. Then, linear Radon transform is put to use to detect the eyelids. Eyelashes are eliminated by threshold.

In normalization stage, “rubber-sheet”model, in general us e, is used t-o unwrap iris image into a rectangle of the same s ize, for the comparison of characteristics.

In feature extraction and coding stage. 1D Log Gabor filter s are used to filter the iris texture features in the space-frequen cy domain. Then, the real part and the image part is quantize d by phase encoding respectively. Besides, the processing of n

oise to generates mask.

In decision training stage, the Hamming distance is employed to measure the similarity of two iris. The minimum distance classifier combining with the threshold with minimum error ratio are used as the decision rules. The experiments implemented on CASIA iris database V3.0 show that the system performs well.

Key Words: Iris Recognition; Iris Location; Circular Hough Transform; Log Gabor Wavelets

目录

第一章绪论 (1)

1.1 生物特征识别 (1)

1.1.1 非生物特征身份鉴别方法 (1)

1.1.2 生物特征识别技术的提出 (2)

1.1.3 研究生物特征识别技术的意义 (3)

1.2 虹膜识别研究 (3)

1.2.1 虹膜识别的机理 (3)

1.2.2 虹膜识别的发展与现状 (6)

1.2.3 虹膜库 (9)

1.3 本文研究的容 (9)

第二章虹膜图像预处理 (11)

2.1 虹膜外边界的定位 (11)

2.1.1 Canny算子和Hough变换的基本原理 (11)

2.1.2 本文采用的Canny算子 (14)

2.1.3 本文采用的Hough变换圆检测算法 (16)

2.1.4 虹膜外边界的定位 (17)

2.2 眼睑的分割 (18)

2.2.1 抛物线检测眼睑 (19)

2.2.2 Radon变换法分割眼睑 (19)

2.3 剔除睫毛 (21)

2.3.1 阈值法剔除睫毛 (21)

2.4 小结 (22)

第三章虹膜图像归一化及特征提取 (23)

3.1 虹膜纹理归一化 (23)

3.1.1 平移 (23)

3.1.2 旋转 (24)

3.1.3 伸缩 (24)

3.1.4 分辨率 (24)

3.2 虹膜的Rubber-Sheet模型 (25)

3.2.1 虹膜边界表示 (26)

3.2.2 虹膜区域的表示 (27)

3.2.3 虹膜区域规化 (28)

3.2.4 灰度级插值 (29)