计算机笔迹鉴别理论与方法

CORC > 自动化研究所 > 中国科学院自动化研究所 > 毕业生 > 博士学位论文

题名	计算机笔迹鉴别理论与方法
作者	刘成林
学位类别	工学博士
答辩日期	1995-06-01
授予单位	中国科学院自动化研究所
授予地点	中国科学院自动化研究所
导师	戴汝为
学位专业	模式识别与智能系统
中文摘要	笔迹鉴别是根据手写笔迹判断书写人身份的一门技术。笔迹鉴别的计算机化是减轻文检人员的工作负担、促使笔迹鉴别技术推广普及的一个重要步骤。计算机笔迹鉴别的研究经历了近30年，但进展并不太大。近年来，社会需要对计算机笔迹鉴别的研究提出了新的要求，模式识别和人工智能等相关学科的进展为计算机笔迹鉴别的发展提供了新的契机。在这样的背景下，本文通过对笔迹鉴别问题的深入分析提出了计算机笔迹鉴别系统研制方案，提出了几种笔迹特征分析和比较以及综合判别的有效技术。论文的主要内容有：介绍了笔迹鉴别的应用背景和发展历史及计算机笔迹鉴别研究的技术状况，分析了笔迹鉴别问题的性质和实现的难点，提出了计算机笔迹鉴别系统的实现方案和研制计划。指出实用的计算机笔迹鉴别系统是人机结合的、机器粗分类和人工专家最终判决的系统，并把计算机自动笔迹特征分析、比较和综合判别方法作为目前研究的重点。在笔迹图象的预处理方面，提出了一种通用的噪声消除方法和有效的字符归一化方法。在特征字的比较过程中，对现有的几种有代表性的方法(高阶相关法、弧模式频率法、方向指数直方图法)进行了实验比较。提出了几种新的字符比较方法：距离变换匹配法、形状特征法、采用Wigner分布的纹理分析方法、多通道分解与匹配法。这些新方法吸收了模式识别和计算机视觉领域的最新理论和技术，在书写入识别和验证实验中取得了很好的效果。这些方法还有一个重要特点，就是没有训练过程，字符比较的距离度量与字符类别无关，从而可以用于书写人和字符不定的情况。在笔迹鉴别领域首次引进了信息融合和证据组合的概念，并用证据组合理论和技术对特征字的比较进行多方法结合和对多个特征字的比较结果进行多证据组合。多方法结合的特征字比较结果比最好的单一方法要好。而结合多个特征字的比较结果对笔迹文件的书写人进行鉴别时得到了很高的正确率。跟现有技术水平相比，本文对计算机笔迹鉴别的理论分析和提出的笔迹特征分析与判别方法具有开拓性，笔迹鉴别的结果有实质性的提高，从而使计算机笔迹鉴别技术向实用化迈进了一大步。
英文摘要	Writer identification (WI) is a discipline which aims to decide the identity of writers according to the handwriting styles. The computerization of WI techniques is the most important step to relax the heavy burden of document examiners and to achieve the goal of prevalent applications. The research of computer WI has a history of near 30 years, which is long in time but achievements in which are not rich. In recent years, social backgrounds urge more achievements in computer WI, and advances in correlated disciplines such as pattern recognition and artificial intelligence supply chances for the development of computer WI. Under this background, This doctoral thesis proposes the overall developing paradigm of cumputer WI via deep analysis of the problem nature, and presents some effective techniques of feature extraction, handwriting comparison and comprehensive decision. The main contents of the thesis are as follows. Firstly the application background and history of development of W1 techniques are briefly introduced. The state of the computer WI researches is also surveyed. The nature of the WI problem and difficulties are analyzed, and then the overall strategy and the plan of building computer WI system are proposed. It is pointed out that an applicable computer W1 system is human-computer interactive, composed of rough classification by computer and final decision by human experts. The automatic handwriting feature extraction, comparison and comprehensive decision is considered the most important researsh tasks of the near target. In respect to the preprocessing of document image, a general noise elimination approach and an effective character normalization approach are presented. In the process of feature character comparison, some existing methods (higher-order correlation, arc pattern based method, and direction index histogram) are implemented and tested comparatively in experiments. And some new methods for character comparison are presented, which are chamfer matching, shape feature method, texture analysis using Wigner distribution, multichannel decomposition and matching. These new methods are motivated by advanced pattern recognition and computer vision methodology, and have achieved promising results in writer recognition and verification experiments. The most important property of these methods is that no training is available and distance measure of character comparison is character class independent. Therefore, they can be used in the case of indeterminate writers and characters. It is the first time that the concepts of information fusion and evidence combination are introduced into WI community in this thesis. The combination of multiple methods for single character comparison and the combination of evidences from multiple characters are implemented by evidence combination techniques from artificial intelligence community. The experimental results of character comparison by multiple
语种	中文
其他标识符	329
内容类型	学位论文
源URL	[http://ir.ia.ac.cn/handle/173211/5648]
专题	毕业生_博士学位论文
推荐引用方式 GB/T 7714	刘成林. 计算机笔迹鉴别理论与方法[D]. 中国科学院自动化研究所. 中国科学院自动化研究所. 1995.