【作者】 陈为民；

【导师】 袁祥辉； 李见为；

【作者基本信息】 重庆大学 ， 仪器科学与技术， 2006， 博士

【摘要】 传统的安全系统主要采用基于信物或口令的方式,随着社会的发展,这种系统显得越来越脆弱。为了应对这些挑战,人们把目光转向了生物识别技术,希望能借助人体的生理特征或行为动作来提高身份识别的精度。在我们的手指表面,因为皮肤的凸凹不平会产生各种各样的纹路,人们常把这些纹路称谓指纹。现代解剖学和统计学已经表明,每个人的指纹都终身不变且与其他人的不一样,同时,每个人各个手指的纹路也各不相同。也就是说,指纹具有稳定性和唯一性。指纹作为主要的生物识别技术之一,被广泛应用于司法、公安和各种安全防护系统。指纹识别作为模式识别最先涉足的领域之一,发展极为迅猛,实际上在很多场合已经成为生物识别技术的代名词。自动指纹识别(AFIS, Automated Fingerprint Identification System)是图像处理技术、模式识别技术与计算机数据库技术的综合应用。它的研究可追溯到上个世纪60年代末,经过近40年的发展,在司法、公安和某些商业领域已取得了成功应用,但目前AFIS仍是模式识别研究的热点之一。针对当前自动指纹识别系统在指纹采集、图像增强、特征表述以及匹配等方面所面临的问题,本文对指纹在增强处理和匹配两方面进行了深入的研究。对于指纹的关键信息获取及校正、图像滤波,特征匹配方面做了大量工作。并对整体算法进行了试验和评价。1、本论文提出并实现了一种基于整体的最优模型残差的方向校正方法,将噪声从指纹整体方向场的拓扑结构中分离出来并进行抑止,比常用的无差别方向低通平滑具有更强的抗干扰能力。2、本论文提出并实现了一种变形的局部数字傅立叶估计频率的方法,并设计了试验进行了验证。该方法能够在较小尺寸的窗口中对指纹的频率进行较高精度的估计,比常用的投影统计以及FFT方法具有较明显的优势。另外,根据指纹区域的主要频率的幅值进行指纹的区域质量评价,为随后的指纹质量自适应滤波奠定了基础。3、本论文提出了指纹的质量自适应滤波。首先从理论上对Gabor滤波器的滤波强度进行了分析。然后根据前面所获得的方向、频率以及区域质量评价确定滤波器参数,对指纹区域采取了质量自适应滤波,使区域质量不同的指纹图像在滤波后达到相对统一的质量,该方法在保护清晰区域的纹线结构和获得劣质区域的特征信息两方面都具有重要的意义,也是本论文算法的核心之一。4、本论文提出了一种以指纹整体特征来描述细节点的方法,并利用指纹方向更多还原

【Abstract】 Traditional security system is mainly based on token or password. With the development of society, this system has become vulnerable. To deal with this challenge, people have turned to the biometrics to improve identity verification by using his physical feature and behavior.A fingerprint is the pattern of ridges and valleys on the surface of a fingertip. Modern anatomy and statistics have proved that each individual has unique fingerprints and fingerprint doesn’t change all life, i. e. fingerprint is steady and exclusive. As a primary biometric, fingerprint has been widely used in forensic, police and other security system and been studied by using pattern recognition technique firstly. Fingerprint identification has been synonymous with biometrics in many situations.Automatic fingerprint identification system (AFIS) has integrated image processing, pattern recognition and database technique. After 40 years of development from1960’s, AFIS has been successful and still the hotspot in pattern recognition domain.Due to the problems in the acquirement, enhancement, character describing and matching, this dissertation is concentrated on two aspects, enhancement and matching. Much work has been done on the key information aquirment and correction, image filtering, feature matching. Experiment and evaluation of the whole algorithm are given.1. In this dissertation, a fingerprint orientation correction based on the optical model residual error is purposed. The noise can be separated from the topology structure of the fingerprint and be removed. Compared with the orientation low pass filtering usually used, our algorithm performance is better in the noise resistance.2. A frequency estimation via improved local Discrete Fourier Transformation and its’verification are proposed. The estimation can be applied on the small scale window with high accuracy. So it is with distinct advantage versus general projective method or the Fast Fourier Transformation. In addition, the fingerprint image quality is evaluated from the value of the master frequency amplitude as the basis of the quality-adaptive filtering.3. A quality-adaptive enhancement is presented. First the filtering intensity of the更多还原