CASIA OpenIR  > 毕业生  > 博士学位论文
指纹匹配与连续分类算法研究
其他题名Research on Fingerprint Matching and Continous Classificaiton
苏琪
学位类型工学博士
导师田捷
2007-06-18
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业控制理论与控制工程
关键词指纹匹配 指纹分类 傅立叶-梅林变换 多项式拟合 分类器融和 Fingerprint Matching Fingerprint Classification Fourier-mellin Transformation (Fmt) Bivariate Polyfit Classifier Combination
摘要指纹识别是生物特征识别领域中应用最广泛的技术,具有悠久的历史。近年来,随着全球经济和信息技术的不断发展,安全问题日益突出,指纹识别技术与电子信息技术相结合发展出的自动指纹识别系统(Automated Fingerprint Identification System, AFIS)越来越广泛地应用到各个方面,譬如个人信息安全领域、电子商务领域、电子政务领域等。 提高指纹识别性能和速度一直是自动指纹识别技术中的两个研究重点。已有的研究成果表明,匹配特征的选取对指纹识别算法性能的影响很大。根据使用的指纹特征,指纹匹配算法可以分为基于局部特征匹配方式和基于全局特征匹配方式。基于局部特征匹配方式主要使用基于FBI细节点模型特征,具有速度快、指纹模板小的优点,但是容易受指纹图像噪声干扰。基于全局特征匹配方式主要使用指纹纹理特征,具有特征稳定、信息丰富的优点,但是匹配精度不高、指纹模板比较大。 针对不同匹配特征的特点,本文从指纹特征提取和表示层次入手,对低质量指纹图像的匹配、连续指纹分类、快速指纹匹配等问题进行了研究,主要研究工作概括如下: 第一,提出了一种基于指纹全局曲率特征的匹配算法。该算法首先沿指纹脊线轮廓线求出指纹曲率初始数据;然后利用二元多项式对指纹全局曲率进行拟合;在指纹匹配过程中使用傅立叶-梅林变换对指纹全局曲率特征进行相关匹配;最终将基于曲率特征的匹配分数与基于细节点特征的匹配分数进行融和,得到指纹模板间的相似度。该算法在FVC2002 和FVC2004数据库上进行试验,结果显示在指纹识别算法中融和基于指纹全局曲率特征可以提高算法的识别性能。 第二,提出了一种基于指纹方向场特征的连续指纹分类算法。该算法首先以指纹参考点为中心将原始指纹方向场信息在复数平面内进行变换提取指纹分类特征;然后使用基于相关距离的计算方法确定分类特征间的相似度。该算法不需要对指纹特征进行旋转配准就能够完成分类特征相似度计算。通过在NIST DB4、FVC2004 DB1数据库上进行测试,试验显示该算法能够较好的对指纹进行分类识别。 第三,提出了一种基于指纹参考点邻近结构的快速匹配算法。该算法以指纹参考点为中心,设计一个包含方向场信息和细节点信息的邻近结构特征。算法利用该结构特征对指纹进行连续分类和匹配,实现大容量指纹数据库的快速识别。该算法在FVC2004 DB1和DB2数据库上进行测试。试验结果显示,该算法在保证自动指纹识别系统识别的准确性的同时,指纹识别速度有一定提高。
其他摘要According the characters of different matching features, this paper focuses on the research of low quality fingerprint matching, fingerprint continual classification and fingerprint fast matching. The main work of this dissertation is as follows: 1) We proposed an algorithm which uses a new representation including global ridge curvature map and local minutiae structure for fingerprint. The ridge curvature feature is represented by the bivariate polynomial and computed by the least square (LS) method. The representation can not only reduce the noise in the ridge curvature map, but also reduce the size of template. We use improved Fourier-Mellin Transformation (FMT) to realize the curvature-based matching. The proposed algorithm combines the score of curvature-based matching and the score of the minutiae-based matching in the decision level by sum rule. We select the FVC2002 and FVC2004 databases to test the algorithm proposed. The experiments prove the combination of two matching features can improve the performance of fingerprint recognition. 2) We propose an algorithm of continuous fingerprint classification based on the combination of ridge width and orientation feature. The algorithm uses two-stage retrieval method to select the candidate fingerprints from the databases. In the first stage, the fingerprint ridge width feature is used to decrease the number of hypotheses from the database. In the second stage, the orientation feature is used to obtain the candidate fingerprints. The orientation feature is extracted by a tessellation structure and transformed into complex mode. Meanwhile, the proposed algorithm utilizes a correlation-based similarity measure. Because of the virtue of the similarity measure, the proposed algorithm does not need the rotation registration process before the feature extraction. Moreover the two-stage retrieval method can efficiently reduce the average search space of the database. The experimental results on the NIST DB4 and FVC2004 DB1 present that the proposed algorithm has better performance than other classification algorithm. 3) We propose a fast fingerprint matching algorithm which combines the classification and matching of fingerprints together, and provide a neighborhood structure, which includes the orientation field and minutia around the reference singular point. This structure has the advantage that the identification information is centralized around the singular point, and can dramatically decrease the calculation amount of matching. It can also be directly used as pattern in both the continuous classification and the fast matching of fingerprints, and carry out the fast identification of the large scale database. Experimental results on FVC2004 databases show that this algorithm can highly speed up the matching of large scale fingerprint database with a preferable performance, and it can be used in the one-to-many matching of the on-line fingerprint identification system.
馆藏号XWLW1124
其他标识符200418014628051
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/6024
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
苏琪. 指纹匹配与连续分类算法研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2007.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20041801462805(2930KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[苏琪]的文章
百度学术
百度学术中相似的文章
[苏琪]的文章
必应学术
必应学术中相似的文章
[苏琪]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。