【作者】 苗晓峰；

【导师】 于明；

【作者基本信息】 河北工业大学 ， 模式识别与智能系统， 2006， 硕士

【摘要】 笔迹鉴别是身份认证的方式之一,在许多领域都有非常重要的应用,目前在公安、社会化考试、银行等领域得到日益广泛的应用,其中一个典型应用就是高等教育自学考试考生试卷笔迹真伪鉴定。目前这些工作还是通过考试中心文检人员手工比对,随着近几年招生人数的增多,这项比对工作,相当耗费人力、物力。本文正是基于高自考这样的背景,来研究基于文本依存的离线笔迹鉴别。本文首先介绍了笔迹鉴别的研究现状和相关理论问题,分析了国内外关于离线笔迹鉴别的研究情况。然后结合课题文本依存的特点,给出一种基于纹理的算法。该算法采用二维Gabor滤波函数对纹理进行分析,选取4个频率8个相位,获得32个核函数,经过变换后,这样对于每一个样本将获得32个变换系数,本文采用32个变换系数的方差作为笔迹特征,在分类时,使用欧氏距离作为分类器,基于同一书写者训练样本少这一特点,本文提出一种新的阈值获取方法。为了对本文算法进行验证,课题在VC++ 6.0环境下编程实现了本文给出的方法,然后采集了30人共108份笔迹样本进行了实验。首先,对每一份笔迹图像进行预处理,得到归一化的训练样本和测试样本;然后对样本使用二维Gabor滤波函数进行快速变换,对变换后的系数求方差,获得32维特征;在训练阶段,使用欧氏距离作为分类器,通过训练样本获取阈值;在测试阶段,将测试数据送入分类器进行鉴别,给出鉴别结果。通过实验得出,正确接受率为89.3%,正确拒绝率为80.0%。更多还原

【Abstract】 Handwriting verification is a method for personal identifications, and it is widely applied in many fields, such as public security, examination, bank etc. The identification of examination papers handwriting is a typical case. At present, the task is completed by hand, and wastes much time and money with the number of students on the increase. So this paper research off line handwriting verification of text dependent based on the background.At first, this dissertation introduces the present research conditions of handwriting verification and related theories, give a survey of handwriting algorithm at home and abroad. Then this paper presents a text dependent method of off-line handwriting verification based on the texture feature. Multi-channel Gabor filter is used in the texture analysis to gain handwriting features. By choosing 4 frequencies and 8 directions, it will get 32 Gabor coefficients, and this paper chooses the variance of coefficients as handwriting feature. The Euclidean distance is applied as classifier, and a new technique for getting threshold value is proposed in this paper.Writer identification programming in this thesis is developed in VC++ 6.0, and this paper collects 108 handwriting samples from 30 persons for the experiment. First, the handwriting samples are pre-processed for getting normalized samples. Second, Gabor filter is used to extract handwriting features, and obtains a 32 dimension vector as handwriting feature. Finally, the Euclidean distance is applied as a classifier for the training and testing process. Experiments show that the correct acceptances rate is 89.3% and the correct rejections rate is 80.0%.更多还原