【作者】 徐露；

【导师】 苑玮琦；

【作者基本信息】 沈阳工业大学 ， 测试计量技术及仪器， 2008， 博士

【摘要】 虹膜识别是20世纪90年代发展起来的一种生物特征识别技术。凭借其高度的准确性,虹膜识别已经得到学术界和企业界越来越广泛的关注。复杂的结构和丰富的特征,使虹膜具有了唯一性、稳定性、可采集性、难更改性、非侵犯性等特点。其中,非侵犯性也是未来生物特征识别发展的必然趋势。本文通过分析虹膜识别系统采集人眼图像的条件,讨论了用于虹膜识别的人眼图像中存在的若干关键问题。针对这些问题,利用CASIA-IrisV3-Interval图像库,对虹膜识别的图像预处理方法以及特征提取和匹配方法进行了研究,试图提出一种具有高准确性的虹膜识别方法。本文的主要工作和成果如下:(1)针对虹膜位移、无关区域、反射光斑等问题,提出了基于人眼图像灰度分布特征的虹膜定位方法:首先,根据瞳孔的灰度分布特征检测瞳孔内一点;然后,利用虹膜内外边界处梯度较大的特点,分别在内外边界上各检测3个边界点;最后,利用不共线的三点可以确定一个圆的原理确定内外边界的圆心和半径。实验结果证明:该方法能够有效地解决虹膜位移等问题,定位成功率为96.12%,平均定位时间为127ms;与经典方法相比,该方法具有更高的定位成功率,却需要更少的定位时间。(2)针对虹膜缩放问题以及眼睫毛和眼睑等干扰问题,利用直角坐标到极坐标的变换,将虹膜归一化成分辨力为512×64的虹膜图像,并在虹膜图像中确定了一个大小为256×32的虹膜有效区域,该区域可以提供无任何干扰的纯净虹膜纹理。(3)针对虹膜有效区域不完整的问题,提出了4种特征提取和匹配方法:首先,提出了基于灰度曲面相似性的虹膜匹配方法,该方法将虹膜有效区域表示为三维空间中的灰度曲面,根据灰度曲面之间的相似性直接实现虹膜匹配;其次,提出了基于局部信息统计的虹膜分块编码方法,该方法将局部信息与全局信息之间以及相邻局部信息之间的比较关系作为可区分性特征进行编码和匹配;再次,提出了基于结构特征的特征提取和匹配方法,该方法提取虹膜纹理的位置特征,并将其表示为二进制的虹膜代码,采用汉明距离对虹膜代码进行分类;最后,提出了基于相位一致性加权平均相位矢量的特征提取和匹配方法,该方法将图像中的每个点在相位一致性最大响应方向上的加权平均相位矢量表示为二进制的虹膜代码,同样采用汉明距离实现虹膜匹配;另外,为了对各种方法的准确性进行客观地评价,将上述4种方法与经典的基于Gabor变换的方法进行了比较。实验结果证明,在本文提出的4种方法中,基于相位一致性加权平均相位矢量的方法具有最高的识别准确性,其正确识别率为98.94%。与经典方法相比,该方法的正确识别率提高了10.63%。综上所述,将该方法作为虹膜识别的核心模块,与图像预处理方法相结合,可以得到一种具有高准确性的虹膜识别方法。(4)针对虹膜旋转问题,在研究灰度曲面匹配方法的过程中,给出了多次平移匹配的方法。实验结果证明,当归一化虹膜图像的分辨力为512×64时,只要令多次平移匹配的次数为12次,即可有效地解决虹膜旋转问题。由于灰度曲面匹配方法应用了图像配准技术的原理,因此该结论对其他特征提取和匹配方法在相同数据库上的研究具有重要的指导意义。更多还原

【Abstract】 Iris recognition is a kind of novel biometrics which was developed from 1990s and it has attracted more and more attention because of its high accuracy.Thanks to its complex structures and abundant features,the iris has many desirable properties,such as unique, stable,collectable,hard to be changed,non-intrusive,and so on.The property of non-intrusive is bound to be the trend of biometrics development.In this thesis,through the analyses of the condition when eye images are captured by the system of iris recognition,some problems contained by eye images used for iris recognition are discussed.In order to resolve these problems,several methods of image preprocessing,feature extracting and matching are researched.Eye images used for experiments come from CASIA-IrisV3-Interval database.Main works and results of this thesis are as follows.(1) Aiming at the problems of iris translation,irrelative regions,and reflected faculaea, method of iris localization based on gray-value distributing characteristics of eye images is proposed.It firstly detects a point in the pupil according to gray-value distribution features of pupil.And it respectively detects three boundary points in the inner and outer boundaries according to the higher gradients of iris boundaries.Then using these six boundary points,it determines the centres and radiuses of inner and outer boundaries according to the principle that a circle can be fixed by three points which are not in the same line.Experimental results demonstrate that the proposed method can effectively resolve all above three problems.The successful localization probability is 96.12%and the average localization time is 127ms.Compared with classical methods,the proposed method has higher successful localization probability but it need fewer localization time.(2) Aiming at the problems of iris flexing and disturbances of eyelashes and eyelids,the iris is normalized to an iris image having the resolution of 512×64 using the transform from Cartesian coordinates to polar coordinates.Then an effective iris region with the size of 256×32 is determined and this region contains pure iris veins without any disturbance. (3) Aiming at the problem that the effective iris region is incomplete,four methods of feature extracting and matching are proposed.Firstly,a method of iris matching based on the similarity of gray-value surfaces is proposed.It represents the effective iris region using a three-dimensional gray-value surface and then it directly achieve iris matching according to the similarity of two gray-value surfaces.Secondly,an iris block-encoding method based on statistic of local information is proposed.It extracts two kinds of comparable relations as distinguishable features to accomplish iris encoding and matching.One is the relation between local and global information,and anther is the relation between neighbor local information.Thirdly,a method of feature extracting and matching based on structural features is proposed.It extracts the position feature of iris veins and it encodes this kind of feature into a binary iris code.Then it achieves iris matching using the Hamming distance.Finally,a method of feature extracting and matching based on weighted mean phase vector of phase congruency is proposed.For each point in the image,the proposed method represents their weighted mean phase vector on the maximal responding orientation of phase congruency into a binary iris code.And then it still accomplishes iris matching using the Hamming distance.In addition,in order to objectively evaluate the accuracy of each method,above four methods are compared with the classical method based on Gabor transform.Experimental results demonstrate that,for these four methods,the method based on weighted mean phase vector of phase congruency has the highest accuracy.Its correct recognition rate is 98.94%.Compared with classical method,its correct recognition rate increased for about 10.63%.Thus,if this method serves as the core module of iris recognition and combines with the method of image preprocessing,an iris recognition method having high accuracy will be obtained.(4) Aiming at the problem of iris rotation,a method of translated matching for several times is proposed in the procedure of researching the method of iris matching based on the similarity of gray-value surfaces.Experimental results demonstrate that,if the translated matching is accomplished for 12 times,the problem of iris rotation can be resolved effectively when the resolution of iris image is 512×64.In fact,the method based on the similarity of gray-value surfaces is a kind of application of image registration,so this conclusion has importantly instructional significance to other methods of feature extracting and matching for the same database.更多还原