摘要
针对现有的文本相关和文本无关的笔迹验证方法无法处理笔迹中仅含有少量且完全不同字符的情况,提出了基于非限定单字字符的内容相关文本无关的笔迹验证方法(半文本无关方法),填补了这一方面的空白。此内容相关文本无关方法通过获取笔迹的字符内容信息,利用线性鉴别分析,可以从不同字符的样本中获取书写入的笔迹风格信息。为了减小不同字符的结构差异所带来的巨大类内变化,引入了表征字符样本内容信息的标准模板,并以内容作因子对原始特征向量与标准模板的差向量进行归一化,最后将这一主要表示字符风格信息的归一化差向量代替原始特征向量用于鉴别分析。实验证明,在笔迹间字符完全不同的极端条件下,该方法比现有的文本无关方法有更高、更鲁棒的性能。
Aiming at the problem that the existing text-dependent and text-independent methods for writer verification cannot deal with the case that handwritings contain a small number of totally different characters, this paper proposes a content related text-independent (semi-text-independent) method to fill the vacancy in the field. Through acquiring the character content information of the handwritings, the content related text-independent method uses the linear discriminant analysis (LDA) to get the writing style information of the writer from the samples of different characters for writer verification. In order to reduce the large within-class variability caused by the structure differences of different characters, this paper first introduces the standard templates representing the content information of the character samples, and then normalizes the difference vectors between the original vectors and the standard templates using content as the factor. Finally, the normalized difference vectors are used to replace the original vectors before LDA. The experimental results demonstrate that, in the harsh condition that the characters between handwritings are totally different, the proposed method has a higher and more robust performance than the existing text-independent methods.
出处
《高技术通讯》
CAS
CSCD
北大核心
2011年第1期1-7,共7页
Chinese High Technology Letters
基金
973计划(2007CB311004)和国家自然科学基金(60772049、60872086)资助项目.
关键词
笔迹验证
内容相关文本无关
线性鉴别分析
归一化差向量
writer verification, content related text-independent, linear discriminant analysis, normalized difference vector