【作者】 高雪莲；

【导师】 杜德生；

【作者基本信息】 哈尔滨理工大学 ， 系统工程， 2009， 硕士

【摘要】 随着信息技术的飞速发展,信息安全已成为当今重要的研究课题之一。目前,传统的安全技术已经不能满足当前的要求,于是人们把目光转向生物识别领域。虹膜识别是一种新兴的生物识别技术,由于其具有唯一性、稳定性、可采集性、非侵犯性等优点而逐步受到人们的重视。虹膜由于其特殊结构,与脸像、指纹、掌纹、声音等生物身份鉴别方法相比,具有更高的准确性。近些年来虹膜识别技术在研究和应用方面都得到了长足的发展,并表现出了广阔的前景和市场。本文讨论了一种新的基于小波变换和奇异值分解相结合的虹膜识别算法。首先,针对已有虹膜定位算法耗时较长和准确率不高等问题,提出一种新的虹膜定位算法:对于虹膜内边界,根据虹膜灰度分布特点,利用灰度平均值法找到瞳孔内初始定位点,并通过边缘检测模板来搜索瞳孔的边界点,同时,引入C-均值动态聚类分析法提高定位精度;对于虹膜外边界,采用由粗到精两步定位的方法。在粗定位确定的虹膜大致范围内,并由虹膜特性分析,采用改进型Daugman微积分算子进行精定位,使用改进的一维参数空间搜索,避免了以往算法在搜索边界时的反复迭代,在保证精确度的前提下,大大提高了定位速度。其次,提出了一种新的基于小波变换和奇异值分解的虹膜纹理特征提取方法。虹膜纹理分布内侧纹理比较密集,外侧纹理较稀疏。根据虹膜纹理分布的特征,将内半圆虹膜图像分成8个图像子块,对每个子块进行三层Bior1.5小波变换提取其中的5个子带并对每个子带奇异值分解降维融合构成最终的虹膜特征向量。最后,在虹膜特征匹配上,给出了改进的加权自适应最小平均距离匹配准则。以CASIA标准虹膜库中的每幅图像作为测试样本,以库中各类聚类特征向量作为该类的特征进行匹配。本文在CASIA虹膜数据库上进行了测试实验,实验结果表明本文提出的算法可靠有效,在识别速度上和识别准确率上都有所提高,识别准确率达到98.36%。更多还原

【Abstract】 With the rapid development of information technology, the research on information security has become one of important topics. However, traditional identification technology is inherently insecure and cannot meet current requirement that leads to a massive rise in the interest for biometric personal identification. Iris recognition is an emerging biometric technology. For uniqueness, stability, available collection and noninvasiveness of iris, iris recognition is being more and more regarded by people. Compared with face, fingerprint, palm-print, voice, and other biometric technology, iris recognition has higher precision. In recent years, iris recognition has made progress in technology research and application, and has a wide prospect and market.A new iris recognition algorithm based on discrete wavelet transform and singular value decomposition is presented in this thesis. Firstly, a new iris localization algorithm is proposed because of the slow speed of localization and inaccurate localization. For inner boundary, initial located point in pupil is gotten by average gray method according to gray distribution. Boundary points of pupil are fixed by the edge detection template, then C-Means dynamic clustering algorithm is brought up to improve precision in localization. For outer boundary, the method combining rude localization with accurate localization is adopted . In the range of rude localization and in light of characteristic analysis of iris, improved Daugman operator is applied for accurate localization in one dimension space. The localization algorithm not only avoids repeated iteration under precision condition but also greatly fastens the speed of localization.Secondly, a method based on discrete wavelet transform and singular value decomposition is discussed to extract iris’s feature. After divided iris into eight small blocks according to its texture distribution, each small block is processed with bior1.5 wavelet transform and 5 strips are extracted.Then singular value decomposition is applied to each strip and the final feature vector is gained.Finally, an improved weighted adaptive criterion of minimum average distance.is raised in feature matching of iris. Each image is chose as testing image and clustering feature vectors in the CASIA iris database is provided for matching by criterion mentioned above.The results of testing experiments indicate that the algorithm is reliable, efficient. The presented iris recognition algorithm is improved both in speed and accuracy, and performs with good recognition rate of 98.36% on the images of CASIA iris database.更多还原