【作者】 刘笑楠；

【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， 2014， 博士

【摘要】 人体虹膜表面具有由坑洞、色素斑、裂缝和环状条纹构成的丰富纹理，这些纹理中所蕴含的特征信息是虹膜身份识别和虹膜诊断的重要依据。现有的虹膜纹理特征提取方法无法获得指定类型虹膜特征纹理的大小、位置、形态和类型信息，而这些信息所反映的虹膜纹理特征是现有虹膜识别系统中特征信息的有效补充，并且能够为虹膜诊断提供所需的纹理特征信息。为了解决上述问题，本文主要研究虹膜图像中特定类型纹理的检测方法。文中依据四种纹理的形状特征将虹膜纹理分为三大类，分别为块状、线状和环状纹理。其中坑洞和色素斑同属于块状纹理，裂缝属于线状纹理，环状条纹属于环状纹理。本文以虹膜块状纹理为检测对象，研究虹膜图像中块状纹理的检测方法、坑洞与色素斑的分类方法以及检测所得块状纹理的特征描述方法。主要研究工作及贡献如下：（1）针对现有虹膜纹理特征提取方法无法获得虹膜中特定类型纹理的大小、形状、位置等特征信息的问题，本文提出一种虹膜纹理特征提取的新思路。将虹膜纹理分为三大类，通过在虹膜图像中检测特定类型纹理，获得其几何和位置等特征信息作为虹膜纹理特征。在虹膜识别领域，这类空间域上的纹理特征信息可以作为现有频域方法所得特征信息的补充。在虹膜诊断领域，可将这些特征作为后续诊断的依据。（2）针对虹膜块状纹理的检测问题，提出三种实现虹膜图像块状纹理检测的方法，这些方法能够实现仅存在色素斑的简单背景虹膜图像和多种纹理并存的复杂背景虹膜图像中块状纹理的检测，三种方法适用环境不同，检测正确率和算法复杂度各具优劣。算法遵循由粗到精的思路，能够依据虹膜图像中是否存在特征纹理将虹膜进行两分类，并且能够在存在块状特征纹理的虹膜图像中实现块状纹理的正确检测，从而得到用于虹膜识别和虹膜诊断的块状纹理特征参数。方法一：基于局部灰度极小值和水平集的虹膜块状纹理检测方法，能够实现仅存在色素斑的简单背景虹膜图像的块状纹理检测。该方法首先根据色素斑的灰度分布特性，在预处理后的虹膜图像中寻找灰度极小值点，确定色素斑存在的可能区域，实现色素斑的初定位。然后，利用水平集方法依据区域内的像素灰度，获得区域图像的边缘轮廓，通过判断所得轮廓的尺寸和闭合性，实现其中色素斑轮廓的检测。测试结果表明，该方法对于仅存在色素斑的虹膜图像中的块状纹理的检测效果较好，检测正确率为92.45%。方法二：基于组合窗口搜索的虹膜块状纹理检测方法，实现了多种类型纹理并存的复杂背景虹膜图像中的块状纹理检测。该方法针对虹膜块状纹理的灰度空间分布和形状特性，设计一组尺寸可变的块状纹理搜索窗口，利用窗口搜索预处理后的虹膜图像，获得满足块状纹理匹配条件的区域，并且排除其它类型纹理。再利用OTSU方法对所得的块状纹理区域进行图像分割，实现块状纹理检测。对图库中人工标定块状纹理的检测正确率分别为90.59%。方法三：基于纹理能量参数与边缘形状因子的虹膜块状纹理检测方法，同样能够实现多种类型纹理并存的复杂背景虹膜图像中块状纹理的检测。该方法先从块状纹理区域的灰度分布特性出发，采用文中定义的区域纹理能量参数作为描述子，利用支持向量机方法将虹膜图像中包含块状、线状和环状纹理在内的灰度变化较剧烈的区域分割出来，实现块状纹理可能存在区域的初定位；再依据块状纹理的形状特征，分析初定位所得区域的边缘信息，针对检测目标的形状特征定义一种边缘形状因子，将块状纹理的轮廓从所得区域边缘图像中提取出来，最终实现虹膜图像中块状纹理的检测。测试结果表明该算法对人工标定块状纹理的检测正确率为85.7%。（3）针对两种虹膜块状纹理的分类问题，提出一种坑洞和色素斑的分类检测方法。该方法首先采用灰度聚类方法实现虹膜图像中块状纹理可能存在区域的初定位，然后对分割所得区域进行闭运算实现相邻区域的连通。再依据坑洞和色素斑的灰度空间分布特性，定义一组由区域灰度标准差和闭运算前后平均变化率组成的区域特征参数作为分类依据，利用支持向量机实现坑洞、色素斑与其它纹理区域的分类。测试结果表明，该方法对于块状纹理的检测和分类正确率分别为87.39%和95.84%。（4）为了获得块状纹理的特征信息，针对块状纹理的特性，选择其面积、周长、长轴长度、短轴长度和重心坐标作为特征参数描述检测所得块状纹理的形态和位置信息，定义了上述参数的计算方法以及虹膜图像块状纹理的特征向量和特征矩阵。另外，文中还讨论了标准分类图库的建立问题与虹膜图像的预处理问题。给出了基于人工分类和统计的不同类型虹膜纹理的标准分类图库的建立方法，并依据彩色虹膜图像的饱和度分量实现了可见光虹膜图像中光照干扰部分的去除。更多还原

【Abstract】 There are plenty of textures on the surface of human irises, which are consisted of crypts,plaques, furrows and circle lines. The feature information contained in iris textures is thesignificant basis for iris identification and iris diagnosis. At present, the existing iris texturesdetection methods cannot offer the detailed characteristics of the iris textures, such as size,location, morphology and category, but these characteristics are the effective supplement ofthe existing feature of iris identification systems and can also provide the texture featurerequired for iris diagnosis systems.In order to solve the problems discussed above, this thesis studied the approaches of iristextures detection for specific type. According to the shape characteristics, four kinds of iristextures were divided into three categories: plaque-like textures, line textures and circletextures. The crypts and plaques were classified as the plaque-like textures, the furrows as theline textures, and the circle lines as the circle textures. Focusing on the plaque-like textures,the thesis studied the detection methods and the classification methods of crypts and plaques.The main contributions of this thesis are as follows:(1) For the problem that the existing iris texture feature extraction methods can not obtainthe features of the specific type of textures, such as the size, shape and position, a new idea ofiris textures feature extraction was proposed. The iris textures were divided into threecategories. By detecting the specific kind of textures, the geometric and located information ofthe iris texture was acquired to use as the iris texture feature. These features of space domaincan supplement the existing frequency methods in the iris identification systems. In the irisdiagnosis systems, these features can be used to get diagnosis.(2) For the problem of the inefficiency of iris plaque-like textures detection, threeapproaches were proposed. These approaches can detect the targets in the iris whether it issimple background only with plaques or complex background with multi-kinds of texturescoexisting. These methods are availiable to different environments and the detectioncomplexity is their merits. Following the coarse to fine ideas, the plaque-like textures existingin the iris image could be detected so that the feature parameters of the plaque-like texturescan be acquired when used in iris recognition systems or iris diagnosis systems. Method I: To solve the problem of plaque-like textures detection of the irises only withplaques, a method based on local gray minimum and level set was proposed. Firstly, accordingto the gray distribution of the plaque-like textures, the local gray minimums were searched forin the preprocessed iris image to find the probable plaque regions. Therefore, the plaques wereinitially located. And then, the regions’ edge contours were evolved with level set method.Finally, by judging the size and closing feature of the contours, the plaques were detectedamong them. The test results show that the method proposed is effective for plaque-liketextures detection of the irises only with plaques, and the accuracy was92.45%.Method II：To solve the problem of plaque-like textures detection of the irises withmulti-kinds of textures, a method based on combinational windows searching was proposed. Aseries of variable size combinational windows was designed based on the gray distribution andthe shape of the plaque-like textures. The preprocessed iris images were searched by thesewindows and the regions in which the plaque-like textures existed were found out and theother regions were excluded. And then, with OTSU method, the regions acquired weresegmented and the plaque-like textures were detected. The integrated accuracy of this methodwas90.59%.Method III: To solve the same problem, a method based on texture energy parameter andedge shape factor was proposed. According to the characteristics of the plaque-like textures, aregion texture energy parameter was defined in this thesis. By using it, the regions containingthe plaque-like, linear or annular textures were detected out by Support Vector Machine, andthe plaques were initially located. By analyzing the edge information of the initialized location,according to the shape of the plaque-like textures, an edge shape factor special for thedetecting target was defined. By using it, the edge contours of the plaque-like textures wereextracted from the edge image, and the plaque-like textures were detected. The integratedaccuracy of this method was85.7%.(3) To solve the problem of plaque-like textures classification, an approach of detectionand classification of crypts and plaques was proposed. First, the probable plaque-like existingregions were initially located by using clustering method based on the gray of the pixels. Andthen, the adjacent regions were connected by close operation. Finally, according to the graydistribution of crypts and plaques, two region feature parameters, i.e. gray standard deviationand average variation rate of close operation, were defined to constitute the feature vector of Support Vector Machine, by which the crypts, plaques and other textures were classified. Thedetection and classification accuracy were87.39%and95.84%respectively.(4) In order to obtain the feature information of the plaque-like textures, according totheir characteristics, the area, perimeter, long axis length, short axis length and barycentercoordinates were selected to be feature parameters to describe the morphology and the positionof the plaque-like textures. The computational formulas of these parameters were proposed,and the feature vector and matrix of iris plaque-like textures were defined.Moreover, the problems of standard classification gallery establishment and thepreprocessing of iris images were discussed in this thesis. The standard gallery of differenttypes of iris textures was established. According to the saturation component of the colored irisimage, the interference of light was removed from the visible light iris images.更多还原