【作者】 回红；

【导师】 汪乐宇；

【作者基本信息】 浙江大学 ， 仪器科学与技术， 2002， 博士

【摘要】 传统的机械钥匙、“口令+密码”以及智能卡等的安全保护措施存在着丢失、遗忘、被复制及盗用的隐患，通过辨识人的生理和行为特征进行身份认证的生物识别技术为其提供了一个方便可靠的解决方案。由于恐怖行动和电脑黑客活动猖獗，近年来生物识别技术越来越多地得到广泛认同。其中，指纹识别技术以方便易用、高准确率和低价格等诸多优势备受关注。虽然有很多指纹系统的商业产品和研究成果问世，但仍不能达到某些特定应用的性能要求。本文在分析了指纹自动识别系统现状的基础上，围绕其核心技术——识别算法展开。针对传统基于细节点的指纹识别方法只利用了指纹图像中的有限信息（细节点）的不足，本文提出了基于指纹脊线纹理结构的特征表达及其匹配算法，冲破了传统方法的局限。 在前三章里介绍了指纹识别技术历史、现状和发展趋势，分析了指纹识别的困难并提出了本文的研究任务；接着介绍了指纹识别系统的组成和评估方法，然后是对指纹识别的一般算法的介绍和剖析。第四章介绍了加以完善的Gabor函数的指纹识别物理模型，描述了从理论设计到实验验证的Gabor滤波器的优化设计方法，对两种格栅方案和两种参数的滤波器分别做了对比实验，确定优化的Gabor滤波器参数；第五章描述了所建立了一整套基于结构的指纹识别算法，采用了一系列改进方法，提高了系统性能；为了缩短滤波时间，第六章提出了指定方向滤波Gabor表达，还描述了基于模糊神经网络（ANFIS）的匹配算法，给出了在指纹识别算法竞赛(FVC2000^|22|)的数据库上的实验结果，并与其他FVC2000算法结果进行了比较：第七章描述的是指纹的补偿以别算法，使识别率大大提高，此方法对指纹识别乃至其他生物识别、图像识别和特征识别技术提供了可以借鉴的依据。 基于结构的指纹识别算法充分利用了蕴含在指纹灰度图像中的丰富的纹理结构信息，又能反映局部信息（细节点），因而弥补了传统方法在小面积指纹的识别上的不足；基于ANFIS的匹配算法和指纹的补偿识别算法，对于指纹图像中的噪声和变形有着很好的鲁棒性，达到了令人满意的识别准确率。所有的算法都经过标准指纹库验证，并与相应的其他方法在误差率方面作了比较，分析了导致算法误差的因素及程度，结出了实验结果和对比，分析了算法的不足之处，并提出了改进方向。更多还原

【Abstract】 Traditional security system,such as mechanical key,password/PIN and passport / ID card,may be forgotten,lost or stolen and cracked by an imposter. The emergence of biometrics has addressed these problems. Biometrics,automatic identification an individual by using certain physiological or behavioral characteristics associated with the person,has gained wide acceptance recently because of the unbridled violence of terrorism and hacker. Fingerprint (FP) recognition is a hot spot and the most widely used branch of biometric with many advantages,such as easier to use,higher reliability and lower cost. A number of commercial systems and research achievements exist for fingerprint recognition;however,the performance cannot satisfy the rigorous requirements of some special application. In this paper,the key technology -FP recognition algorithm is launched based on the analysis of current situation of automatic fingerprint identification system (APIS). The structure-based FP representation and its matching algorithm are proposed to address the shortcoming of traditional minutiae-based method,which makes use of limited information in FP images.In the former three chapters,we introduce the history,status quo and the development trend of APIS,the difficulties of FP recognition,and the object our research;also the combination and access method for the APIS;moreover,the anatomy of traditional method for FP recognition.In the forth chapter,the physics model of Gabor function for FP recognition is present. The optimal designed Gabor filters is illustrated theoretically and experimentally. The comparing experiments for two tessellated methods and two parameters selections are given for confirm the optimal filters.In the fifth chapter,the whole FP recognition algorithm is described in details,which made much improvement to enhance the performance.In the sixth chapter,the orientation specified filtered representation is described to reduce time consume of filtering operation,and then the ANFIS-based (Automatic Neuro-Fuzzy Identification system) matching algorithm is proposed. The experimentresults compared with other methods on the standard FP database (FVC2000 [22]-FP Verification Competition) are given.The compensated recognition algorithm is proposed to improve the recognition rate in the seventh chapter. This method could be used for reference by other FP technology,and also by other biometric,image and pattern recognition and classification,etc.The main conclusion,the shortage and the prospect of the research of this thesis are given finally.Structure-based FP recognition method not only makes best use of texture structure information contained in the gray level FP image but also reflect the local information. So it can offset the shortage of traditional method in limited area image. ANFIS-based matching algorithm and compensated recognition algorithm are robust to noise and distortion,and achieve a high accuracy. All of the experiments are tested on standard FP database,and compared accuracy with other related methods. The element and grate of what can result in error is analyzed and compared by experiments.更多还原