自动指纹识别系统研究

【作者】 彭章平；

【导师】 彭小奇；

【作者基本信息】 中南大学 ， 电路与系统， 2007， 硕士

【摘要】 生物识别技术是依赖人体的生物特征进行身份验证的一种解决方案。每个人的指纹具有唯一性且终身不变，因此指纹识别是代替传统身份识别手段的最安全、最可靠、最方便的方法之一，是比较理想可行的一种生物识别技术。本文简要介绍了生物识别技术和指纹识别系统(AFIS)的结构与工作原理，分析了指纹图像预处理、特征信息提取、指纹分类和匹配识别技术的发展历史及其研究和应用现状，在对国内外当前流行的一些指纹预处理、指纹特征提取算法进行深入研究和分析的基础上，取得了如下的研究成果：1．提出了一种基于灰度特性的指纹前、背景分割方法。该方法能消除分割过程中出现的孤立块，对噪声干扰有一定的鲁棒性，实用性较强。2．在对基于固定方向滤波的指纹图像增强算法及其效果进行研究分析的基础上，提出了一种改进的Gabor滤波指纹图像增强算法。该方法根据指纹纹线的结构特点，通过对纹线投影信号进行频谱分析来准确计算指纹的纹线频率。实验表明，该方法对纹线粘连图像有很好的处理效果。3．把指纹图像看作一种局部平稳而总体非平稳的信号，提出了一种基于统计学的指纹方向和频率估计算法。实验结果表明，该算法对指纹图像有显著的增强效果，并有效地提高了指纹细节特征提取的准确率。4．针对指纹图像噪声的特点，设计了一种能滤除指纹图像中的冲击噪声的方向中值滤波算法。该算法采用局部阈值法二值化指纹图像，有助于保护纹线的局部信息，避免了图像二值化过程中引入过多的误差。实验结果表明，该方法效果良好。5．针对图像预处理过程产生的伪特征点难以消除的问题，提出了一种基于信息融合的特征点提取方法。通过融合两种不同预处理方法所获得的结果来初步筛选特征点，然后根据特征点的结构信息来消除伪特征点。实验结果表明，该方法不仅可以有效地消除因预处理而产生的伪特征点，同时也能消除因图像采集质量等原因而产生的伪特征点。更多还原

【Abstract】 Biometrics technique is the automatic identification of a person based onhis/her physiological or behavioral characteristics. Due to the uniqueness andinvariability of fingerprints, the fingerprints identification is becoming a mostsecure, reliable, convenient alternative to the traditional identification methods,and it is an ideal and feasible biometrics technique comparatively. This paperdescribes the structure and working principle of the biometrics technique andAutomatic Fingerprint Identification System (AFIS), analyses the developmenthistory and the application, investigates status of fingerprint imagepreprocessing, minutiae extraction, fingerprint classification and matching, andprovides an in-depth study and analysis on the current popular of fingerprintsimage preprocessing, and fingerprint feature extraction algorithm. The mainresults for the research are as follows:1. A fingerprint segement method based on gray level statistics is proposed,which can eliminate the isolated pieces arised from the segmentation process,with robustness to noise and practicability.2. An improved Gabor-based fingerprint image enhancement method isproposed from the research of image enhancement algorithm and effect forfixed direction filter. According to the fingerprint ridge characteristics, the ridgefrequency is accurately calculated via ridge projection signal spectrum analysis.Experiments show that the method has good effect in ridge eonglutination area.3. Taking fingerprint image as a signal which is locally stationary butnon-stationary as a whole, the paper puts forward a probabilistic approximationmethod for the ridge orientation and frequency. Experimental results indicatethat the method has prominent effect on fingerprint images, and improves theaccuracy of minutiae extraction.4. According to the fingerprint image noise characteristics, a directionmedian filtering algorithm is designed, which can filter the impulse noise in thefingerprint image. The algorithm uses local threshold to binarize the fingerprintimages, contributes to the protection of the ridge local information, and avoidsintroducing too many errors in the binarization process. Experimental resultsprove that the algorithm is quite effective. 5. A minutiae extraction method based on the information fusion isproposed in order to solve the problem that the pseudo minutiae arised from theimage preprocess is diffcult to eliminate. The method filtrates the minutiae byfusing two results of different preprocess methods, and then eliminate pseudominutiae according to its structure. Experimental results show that the methodcan not only effectively eliminate the pseudo minutiae arised from the imagepreprocess, but also eliminate the pseudo minutiae caused by image acquisitionor quality et al.更多还原