【作者】 刘菲；

【导师】 尹义龙；

【作者基本信息】 山东大学 ， 计算机科学与技术， 2015， 硕士

【摘要】 随着信息化时代的到来和人们虚拟活动空间的不断增大,人们对个体身份识别技术的要求越来越高。如何高效准确的进行个体身份鉴别、保护信息安全是当今时代急需解决的关键社会问题之一。传统的身份识别技术早已满足不了当前实际的应用需求,新时代要求我们必须寻找一种新的途径来更加安全方便的进行身份识别。因此基于生物特征的身份识别技术,由于其高安全性、高稳定性等特质,引发了诸多海内外研究者的兴趣,其中的人脸识别技术、指纹识别技术已趋于成熟。而新兴起的手指静脉识别技术,因其拥有的诸多优势得到了越来越多研究者的青睐,具有良好的应用前景。目前,手指静脉识别的主流方法是基于像素级低层次特征的方法,所使用的特征主要是基于图像整体区域提取到的特征以及基于静脉纹路区域提取到的特征。其中基于图像整体特征的方法会存在大量的冗余数据,时间复杂度较高,而基于静脉纹路特征的方法则容易受图像分割算法性能的制约。为了提高静脉特征的有效性,提高身份识别的效率,本文提出了一种新的基于细节点特征的手指静脉识别方法,并利用奇异值分解算法进行细节点配对。此外,为了尽量消除单个像素点噪声对性能的影响,提高系统的识别性能,本文还提出了一种基于超像素特征的手指静脉识别方法。本文首先从方法论的角度对文中涉及的一些基础性算法,包括超像素过分割,奇异值分解以及空间金字塔匹配核等进行了概述；其次分别从研究动机,到方法流程,再到实验结果及分析,详细介绍了本文所提出的两种新的手指静脉识别方法——基于奇异值分解的手指静脉识别方法和基于超像素特征的手指静脉识别方法；最后,对全文的各项工作进行了总结,并对未来的研究重点进行了阐述。基于奇异值分解的手指静脉识别方法实际是一种基于细节点特征的手指静脉识别方法,该方法首先通过奇异值分解一个加权相关邻接矩阵来获取细节点匹配对；其次通过评估上述匹配对的相似度来去除伪匹配对；最后通过平均各组匹配对的相似度得分来完成图像匹配得分的计算。基于超像素特征的手指静脉识别方法实际是一种基于高层次特征的手指静脉识别方法,该方法首先使用某种单像素特征提取算法提取图像的基特征；其次某种采用超像素过分割技术获得图像的超像素块；之后基于基特征,利用统计技术提取超像素特征；最后使用加权空间金字塔匹配算法计算匹配得分。在不同数据库上进行的大量实验表明,本文所提出的基于奇异值分解的和基于超像素特征的手指静脉识别方法均具有良好的识别性能及应用价值。在后续工作中,我们将把重心放在手指静脉新特征的设计和使用,以及新匹配算法的研究上。更多还原

【Abstract】 In modern society, with the rapid development of information technology and the continuous extend of human physical and virtual space, human identity recognition is be put forward higher requirements for safety and practicality. One of the most major characteristics of rapid development of information technology is the digital and hidden of the personal identity. Thus, how to identify a person’s identity accurately and protect the information security is a key social problem in today’s information age.Traditional identification methods can’t meet the requirements above. Human beings must seek a new way of identification with high security and convenience. Thus, the identity recognition based on biometrics has attracted more and more researchers’ attention. Biometrics refers to verify identity using some physiological or behavioral characteristics of human body itself. Compared to traditional biometrics such as fingerprint recognition, finger vein recognition is a new biometric technology which has some better prospects.Currently, most of the finger-vein recognition methods are based on pixel-level features. These low-level features are mainly extracted on the whole image or the finger-vein networks. And they have some problems. The finger-vein recognition methods based on the whole image will get a lot of redundant data and their time complexity is high. The finger-vein recognition methods based on the finger-vein network will be sensitive to the image segmentation. Thus in order to improve the effectiveness of the features so as to improve the recognition performance, in this paper, we propose a new finger vein recognition method based on singular value decomposition (SVD). In addition, in order to eliminate the impact of the noise onthe single pixel, we used superpixel feature information and propose a novel method.This paper firstly gives the overviews of the key algorithms in our work, including the superpixel over-segmentation, the singular value decomposition, and the spatial pyramid matching kernel, besides, introduces their research background, algorithm flow and application; secondly, respectively from the research motivation, process, and the experimental results, it introduces two new methods. One is the SVD-based finger vein recognition and the other is superpixel-based finger-vein recognition; finally, we summarize the work in this paper and prospect the follow-up work.The SVD-based finger vein recognition method involves three stages:(Ⅰ) minutia pairing, (Ⅱ) false removing and (Ⅲ) score calculating. In particular, stage Idiscovers minutia pairs via SVD-based decomposition of the correlation-weighted proximity matrix. Stage Ⅱ removes false pairs based on the local extensive binary pattern (LEBP) for increasing the reliability of thecorrespondences. Stage Ⅲ determines the matching score of the input and template images by the’average’.The superpixel-based finger vein recognition method utilizes superpixel-based features of finger vein for high level feature representation. When comparing two finger veins, the features of each pixel are firstly extracted as base attributes. Then, after superpixel over-segmentation, the SPF of eachfinger vein can be obtained based on its base attributes by some statistical techniques. Lastly, a weighted spatial pyramid matching scheme is utilized to implement matching.Extensive experiments demonstrate that the methods notonly perform better than the similar works in the literature, but also have great potential to achievecomparable performance to other categories of state-of-the-art methods.For future extensions of our work, we are interested inimproving the accuracy of the minutiae extraction algorithm usedin this paper and utilizing some weighted fusion methods forfeature combination.更多还原