【作者】 王浩；

【导师】 苑玮琦；

【作者基本信息】 沈阳工业大学 ， 精密仪器及机械， 2009， 硕士

【摘要】 虹膜识别被认为是目前一种最可靠并且准确率最高的生物特征识别系统。近几年,一些机场已经开始安装虹膜识别检测系统,很多公司也在不断开发新一代的虹膜识别装置。但是,对于虹膜识别技术的大规模使用仍然存在着一些技术上的问题。在虹膜识别过程中,人以正常步行速度通过采集装置会出现非理想的虹膜图像,其中包括角度偏离虹膜、离焦虹膜、运动模糊虹膜。本文试图解决角度偏离虹膜的识别问题。如果要保证非理想虹膜也能正确的识别,关键是能否在人眼图像中准确地定位到虹膜。本文提出了一种基于椭圆投影的定位算法,该算法利用最小二乘法拟合椭圆形虹膜内边界,并依据椭圆与圆形的近似度检测人眼凝视方向,应用投影变换把角度偏离的虹膜转换成正视角度,三点定位圆的方法定位虹膜外边界。然后就可以用传统的特征提取与匹配算法对校正后的正视角度虹膜进行处理。为了更有效地提取虹膜纹理,本文提出了两种特征提取算法:一种是基于二维不可分小波相关性分析的特征提取算法,其中用到的二维不可分B-样条小波变换比传统结构的二维小波变换具有更多的方向,并且能够快速收敛成高斯函数。应用这种小波结构后,识别算法的识别效果与Daugman的二维Gabor方法相近,并且它是离散化的滤波器,滤波速度明显快于连续的二维Gabor。因为小波域特征的相关性与识别效果成正比,可以根据相关性大小对于得到的小波系数特征选择最优的子特征。实验表明,该算法比经典的虹膜识别方法能更准备地捕捉识别效果好的特征区域。另一种是基于非下采样Contourlet变换(NSCT)最大响应方向的虹膜特征提取算法。非下采样Contourlet变换是多方向的第三代小波变换,由于没有下采样操作,使之具有平移不变性。该算法应用非下采样Contourlet变换对虹膜进行特征提取,并根据虹膜纹理特点分块选择响应方向最大的特征进行匹配。实验表明,该算法具有非常优越的识别性能。更多还原

【Abstract】 Iris recognition is regarded as the most reliable and accurate biometric identification system available. Recent years, some airports have started to install iris recognition system, and multiple companies are developing a new generation iris recognition products. However, there are still some technical problems about large-scale use of that technology.There will be non-ideal iris images when the person is moving at a normal walking pace through a iris capture device, such as off-angle iris, defocus iris, motion-blurred iris. We are trying to solve the identification problem of off-angle iris. If non-ideal iris can be correctly identified, the key point is whether the iris will be located in the human eye image accurately. In this thesis, a non-ideal iris localization method based on elliptical projection is proposed. Firstly, this method uses the least squares to fit elliptical iris inner boundary. Secondly, eye gaze direction is detected by the similarity of ellipse and circularity, and an off-angle iris is converted into a frontal view iris using projection transform. Thirdly, iris outer boundary is located by the method of three points fitting a circle, and rectified iris image is processed using traditional feature extraction and matching algorithms.In order to extract the iris texture efficiently, two feature extraction algorithms are proposed: the first one is iris recognition based on correlation of 2-D nonseparable wavelet. 2-D nonseparable B-spline wavelet transform used in this thesis has the property that multiple directions and rapidly converge to a Gaussian as the order increases. Therefore, the used wavelet structure has the similar recognition performance with Daugman’s 2-D Gabor method. Besides, that wavelet structure is the discrete filter, so the filtering rate is significantly faster than a continuous 2-D Gabor. Because the correlation of wavelet domain features is proportional to recognition performance, the optimal features can be selected from wavelet coefficients according to the value of correlation. Experiments show that the algorithm proposed can capture feature areas of good recognition performance more accurately than the classical iris recognition algorithms. The second one is iris feature extraction algorithm based on nonsubsampled contourlet transform (NSCT) with maximal responding direction. NSCT is the third generation of wavelet transform with multi-direction. Because of its nonsubsampled structure, NSCT is shift-invariant. NSCT is used to extract feature from iris, and according to characteristic of the iris texture, feature blocks with maximal responding direction is selected to match. Experiments show that algorithm has superior recognition performance.更多还原