基于局部轮廓方向特征的笔迹鉴别

Writer Identification Based on Local Contour Direction Features

利用局部结构的概率分布能够体现笔迹特征的思想,提出一种基于局部轮廓方向特征的离线笔迹鉴别方法。使用与窗口中心连通的边缘片段减少笔迹粗细的影响,将边缘点分成32种方向类别从而获取辨别局部轮廓的能力。在此基础上,统计以边缘点为中心的滑动窗口中的方向特征并归一化为局部轮廓方向特征。最后,使用加权Manhattan距离对笔迹特征进行相似性度量。实验表明提出的方法在ICDAR2011笔迹鉴别竞赛的多语言测试数据库上获得了较好的辨别正确率,部分指标超过了目前的先进算法。

A method based on local contour direction feature is proposed for off-line writer indentifaction in this paper. It reduces the inpact of stroke wideth by ignoring the fragments which do not directly connect the center point. The edge points are divided into 32 catigories to gain the directions of local fragments. This method counts the direction features in sliding windows and normalizes them into local contour direction features. The weighted Manhattan distance is used as similarity measurement at last. The experiments on ICDAR 2011 writer identification database with multi-languages show that the performances of the proposed method reach or exceed the state-of-art methods.