视网膜血管识别技术研究与算法实现

发布时间：2018-09-03 11:37

【摘要】：随着电子商务、手机支付和在线购物等线上交易模式的发展,人们对于虚拟数据的安全性的要求越来越高,传统的在线身份识别技术,如账号密码已不能满足安全性的需求,而生物识别技术得益于其可靠性、难伪造性开始逐渐被应用到手机、电脑等民用电子终端设备的身份认证和支付系统中。在众多生物特征之中,视网膜是最可靠、最稳定以及最难被伪造的生物特征之一,因此它非常适合作为身份认证的标识。在可预见的未来,视网膜识别技术有很大的希望被应用到在线支付、访问权限控制、自动取款等对安全性需求高的民用领域。所以对于视网膜识别技术的研究有着极大的价值和良好的前景。本文深入研究了大量与视网膜和生物识别相关国内外论文,特别是在视网膜图像的血管分割、特征提取以及特征匹配方面。然后对这些论文提及的相关算法的进行挑选、整合和改进,设计了一套准确稳定的视网膜识别方案。本文的主要工作和成果如下:(1)在视网膜图像分割阶段,研究了多种血管分割的算法,并加以对比。结合视网膜血管的特征,最后采用高斯滤波对视网膜血管进行增强,并使用最大二维熵阈值算法进行血管分割。通过实验检验,该方法对血管分割的效果要优于其他常用的边缘分割算法并具有良好的抗噪性。(2)在特征点提取阶段,根据对视网膜血管的研究,选用血管的分叉点作为特征点,并利用形态学的知识对图像做了细化处理。然后对Harris角点特征提取算法和邻域特征提取算法作了详细的分析,并将它们的提取结果标注出来进行对比,确认领域特征提取为最优的针对细化图的提取方式。除此之外,还对利用二次特征提取来进一步排除伪特征点,增加了有效点的比例,提高了提取的准确度。(3)在特征点匹配阶段,研究和对比了几种常用的特征匹配方法,从稳定性、准确度和效率等多个方面衡量了它们的优缺点。通过分析和比较,最终设计出了一种融合算法,将三角形匹配算法和二维聚类算法结合起来,充分利用了两者各自的优点,使匹配算法同时具备稳定性和效率。除此之外还对模板三角形的选取方法和相似三角形检索算法进了一定的改进,提升了匹配的效率。
[Abstract]:With the development of electronic commerce, mobile phone payment and online shopping, people are demanding more and more security of virtual data. Traditional online identification technology, such as account password, can no longer meet the need of security. Because of its reliability, biometric technology has been gradually applied to the identification and payment systems of mobile phones, computers and other civil electronic terminal devices. Among the many biometric features, retina is one of the most reliable, stable and difficult to be forged, so it is very suitable for identity identification. In the foreseeable future, retinal recognition technology has great hope to be applied to online payment, access control, automatic withdrawal and other civilian areas with high security requirements. Therefore, the research of retinal recognition technology has great value and good prospects. In this paper, a large number of papers related to retina and biometrics have been deeply studied, especially in vascular segmentation, feature extraction and feature matching of retinal images. Then we select, integrate and improve the relevant algorithms mentioned in these papers, and design a set of accurate and stable retinal recognition scheme. The main work and achievements are as follows: (1) in the phase of retinal image segmentation, a variety of blood vessel segmentation algorithms are studied and compared. According to the characteristics of retinal vessels, Gao Si filter is used to enhance retinal vessels and the maximum two-dimensional entropy threshold algorithm is used to segment the retinal vessels. The experimental results show that this method is superior to other commonly used edge segmentation algorithms and has good noise resistance. (2) in the phase of feature point extraction, according to the research of retinal vessels, the bifurcation points of blood vessels are selected as feature points. The image is thinned by morphological knowledge. Then, the Harris corner feature extraction algorithm and the neighborhood feature extraction algorithm are analyzed in detail, and their extraction results are compared to confirm that the domain feature extraction is the best extraction method for thinning image. In addition, we use quadratic feature extraction to further eliminate pseudo-feature points, increase the proportion of effective points, and improve the accuracy of extraction. (3) in the phase of feature point matching, several common feature matching methods are studied and compared. Their advantages and disadvantages are measured in terms of stability, accuracy and efficiency. Through analysis and comparison, a fusion algorithm is designed, which combines triangle matching algorithm and two-dimensional clustering algorithm, and makes full use of their respective advantages, so that the matching algorithm has the stability and efficiency at the same time. In addition, the selection method of template triangle and the similar triangle retrieval algorithm are improved to improve the efficiency of matching.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R770.4;TP391.41

【参考文献】