【作者】 孙晓琳；

【导师】 郭树旭；

【作者基本信息】 吉林大学 ， 电路与系统， 2012， 硕士

【摘要】 随着社会经济的蓬勃发展，人们对信息安全的要求越来越高，传统的身份认证方式由于其无法满足人们日益增长的对身份识别的需求，而渐渐退出了历史舞台，取而代之的是新一代的生物识别技术。指纹识别技术作为生物识别技术中的重要一员已经有了较为成熟的发展，但是由于受其自身的安全性，稳定性方面的因素制约，目前的发展已经进入了瓶颈阶段。近年来手指静脉识别技术作为指纹识别技术的替代品逐渐得到了人们的认同。指静脉识别技术较指纹识别技术有着更好的稳定性：指静脉成像不会受到人体外伤，磨损等因素的影响；同时也有着更高的安全性：由于静脉是人体内部的特征，使妄想伪造指静脉的不法分子没有了可乘之机。正是由于它的一系列优点，使指静脉识别技术成为了学者们研究的新宠。本文围绕指静脉图像模式提取算法进行了一系列的研究。本文首先介绍了指静脉图像的获取，由于血液中的失氧血红蛋白能够吸收近红外光线，所以手指静脉能在近红外光下成像。然后本文使用四种经典的图像分割方法对指静脉图像进行分割，实验表明，处理效果并不是很理想。于是本文研究使用最大曲率算法对指静脉图像进行模式提取，从静脉位置剖面图的五个角度对图像进行处理，最终得到了很好的实验结果。本文对应用最广泛的Canny边缘检测算法进行了改进，使用双边滤波器代替Gauss滤波器，在不损失信息能量的基础上使去噪更彻底，得到了更好的边缘检测结果。本文还对指静脉图像的合成算法进行了深入的研究，总结了真实指静脉图像中静脉的分布规律，依照这些规律使用最小二乘法对指静脉曲线进行拟合，然后与合成的静脉背景模式进行融合，得到合成指静脉图像。最后文章对指静脉图像的特征提取进行了研究，验证Hu矩的不变性并使用Hu不变矩的理论对图像进行特征提取。本文的研究内容主要围绕以下方面展开：第一部分，文章介绍了指静脉识别技术的研究意义，国内外的研究现状与市场发展状况。指出指静脉识别技术是继指纹识别技术之后，有着广泛发展前景的一项新研究领域。最后对本文的研究内容，文章框架进行了介绍。第二部分，文章介绍了指静脉图像的获取方法，指出指静脉在近红外光线的照射下容易成像，因为血管中的血红蛋白容易吸收这个波长范围内的光线。使用近红外光线对手指进行扫描，采集得到的指静脉图像。首先需要完成的一步是对图像进行预处理，预处理主要包括：调整图像直方图，使直方图均衡化；归一化图像灰度等。便于对图像进行进一步的处理。本文分别使用四中经典的阈值分割算法对指静脉图像进行分割：固定阈值法，总体均值法，最大类间方差法，阈值图像分割法。实验结果表明以上四种算法均不能对指静脉图像进行有效的处理。原因是静脉图像的灰度对比度不强，图像中含有大量噪声，造成图像质量较低。第三部分，使用最大曲率算法分割图像，提取指静脉中心线。首先介绍最大曲率算法的基本原理，对指静脉图像取位置剖面图，经分析发现静脉部分的灰度值由静脉两侧向静脉中心部分递减，直至中心线部分达到局部最小值，遂决定使用曲率的思想搜索静脉中心点。本文分别从0°，30°，60°，120°，150°五个方向上对静脉取位置剖面图，分别处理后对中心点进行汇总，最终得到静脉中心线，经细化和骨架提取之后完成静脉模式的提取。第四部分，本文使用双边滤波器对Canny边缘检测算法进行改进。首先对传统的Canny边缘检测算法原理进行介绍，首先进行Gauss滤波，其次非极大值抑制，最后采用滞后阈值法连接边缘。本章提出值域滤波器和定义域滤波器的概念，使二者结合得到双边滤波器，双边滤波器能在不损失信号能量的基础上更彻底的去除噪声。本文使用双边滤波器取代Canny边缘检测算法中的Gauss滤波器，使用倒高斯模型提取静脉中心线。第五部分，对静脉合成算法进行了深入的研究。首先对真实的指静脉图像进行分析，总结了静脉分布的规律，依据规律使用最小二乘法对静脉曲线进行拟合，得到基本静脉模式。然后累加100幅真实指静脉图像，取其平均值作为合成指静脉图像的背景模式。使用加强系数和衰减系数作为其权值，融合静脉模式和背景模式。通过调整参数，可以模拟多种不同情况下的静脉模式。合成指静脉算法的提出对静脉处理算法的评价、优化有着极其重要的意义。第六部分，对指静脉图像的特征提取进行了研究。本文使用Hu不变矩的理论对指静脉图像进行特征提取，可以避免由于压缩，偏移，光照不均等因素对静脉成像造成的影响。首先对Hu矩进行验证，证明了Hu矩具有伸缩、平移、旋转的不变性。然后利用Hu不变矩对静脉进行特征提取。第七部分，对全文进行总结，对下一阶段的工作进行展望。更多还原

【Abstract】 With the vigorous development of the socio-economy, the requirements of people forinformation security are increasingly high. Due to the disability to meet the growingdemands for identification, the traditional authentication methods gradually withdrawfrom the stage of history, replaced by a new generation biometrics. The fingerprintrecognition technology as an important member of biometric technology has a maturedevelopment. But subjecting to a constraint upon its own security and stability, thecurrent development has entered a bottleneck stage. In recent years, finger-veinrecognition technology as a substitute of fingerprint recognition technology, graduallygains people’ recognition. Compared to fingerprint recognition technology, the veinrecognition has a better stability: the imaging of finger-veins would not be affected byhuman traumas, wears and other factors. It also has a higher security: veins are thebody’s internal characteristic, so there is no opportunity for criminals to forge theveins. It is the series of advantages that make the finger-vein recognition technologythe new darling of the scholars’ research.This paper does a series of studies focusing on the pattern extraction algorithms offinger-veins. This paper first introduces the acquisition of finger-veins images.Because the deoxyhemoglobin in blood can absorb near-infrared light, the finger-veins can be imaged under the near-infrared light. This article uses four classic imagesegmentation methods to segment finger-veins image, and the experiments show thatthe treatment effect is not very satisfactory. Consequently this paper uses thealgorithm of maximum curvature to extract the patterns of finger-veins, whichprocesses an image from five angles of its section where the vein is located and finallygets a good experiment result. In this paper, the most widely used Canny edgedetection algorithm is improved by using bilateral filter instead of the Gauss filter. Itwill denoise more thoroughly and get a better edge detection result on the basis of noloss of information energy. The synthesis algorithm of finger-vein images is alsoresearched deeply. The distribution laws of the veins in real image are summarized,according to which the vein curves are fitted, with the method of least squares. Andthen these vein curves are syncretized with the synthetic vein background mode to getthe synthesis finger-vein image. At last, this article studies the characteristics extraction of the finger-vein image. The invariance of Hu moments is verified andused to extract the features of vein images.The study of this article is mainly focused on the following areas:In the first part, this article describes the significance of finger-vein recognitiontechnology, the research status of domestic and international and the marketdevelopment. And then points out that the vein recognition technology has a broadprospect for development following the fingerprint recognition. Finally, the mainstudy contents and the framework of this paper are introduced.In the second part, the acquisition method of finger-vein image is described, referringthat the vein is easy to image in the near-infrared irradiation which because thehemoglobin in the blood vessels is easy to absorb the light in this wavelength range. Aseries of pre-treatments is done after obtaining the finger-vein image, includingadjustments to the image’s histogram to make it balanced, and normalization the grayof image to facilitate further processing of the image. Four classic thresholdsegmentation algorithms are used to segment the image: the fixed threshold method,the overall average method, the OTSU method and the threshold image segmentationmethod. The results of experiments show that the above four algorithms can not dealwith the finger-vein image effectively. The reason is that the gray-scale contrast ofvein image is not very strong and there is a lot of noise in the image, resulting in thedescent of image quality.In the third part, this article uses the maximum curvature method to process the imageaccording to the poor quality of vein image. At first, the basic principles of maximumcurvature algorithm are introduced. The location profile of finger-vein image isobtained. The analysis find out that the gray value of the vein declines from the bothsides to the central part of the vein, and it will reach the local minimum at the centerline. Then the concept of curvature is decided to use to search for the center point ofvein. The position profiles are got from five directions which are0°,30°,60°,120°and150°. After each of them is processed respectively, the center line is obtained bygathering the five directions’. The complete vein pattern is extracted after skeletonand refinement.In the fourth part, the Canny edge detection algorithm is improved based on thebilateral filter. The processes of traditional Canny edge detection algorithm are firstintroduced: filtering by Gauss filters, followed by non-maxima suppression andfinally the method of hysteresis thresholding is used to connect the edge. This paper proposes the concept of range filters and domain filters. By combining those twofilters, the bilateral filter is obtained which is able to remove the noise without loss ofsignal energy. This article uses bilateral filter to replace the Gauss filter in the Cannyedge detection algorithm, and uses an inverted Gaussian model to extract the veincenterline.The algorithm of synthesis finger-vein image is introduced in the fifth part. First, thedistribution laws of real finger-vein image are summarized, in accordance with which,the vein curves are fitted using the least squares method. And then the100real finger-vein images are accumulated. The mean value is used as the background mode ofsynthesis finger-vein image. With strengthen coefficients and attenuation coefficientsas weights, vein patterns and background mode are syncretized. By adjusting theparameters, it can simulate the vein pattern under a variety of circumstances. There isa great significance for the evaluation and optimization of vein processing algorithms.In part six, the feature extraction of finger-vein image is studied. This article uses thetheory of Hu invariant moment to extract the features of finger-vein image, which canavoid the influences caused by compressions, offsets, uneven illuminations and otherfactors on venous imaging. First, the validation is done on the Hu moments to provethe invariance of stretching, translation and rotation. Then the Hu invariant momentsare used to extract the features of finger-vein.The seventh part summarizes the paper, and looks ahead to the next phase of work.更多还原