摘要
针对手写笔迹的高度相似性与随机性提出一种利用几何矩定量表示字符形状特征的笔迹鉴别方法。该算法先搭建文本依存的词袋模型,提取几何矩的八个不变特征。在分类决策层利用因子分离与特征融合方法,实现文本独立的笔迹分类。该算法在IAM数据集中的首选(Top-1)鉴别率已达到96.7%,取得了同类研究中前四的成绩。实验结果表明,基于词袋模型和矩特征的笔迹鉴别方法在维吾尔文以及英文数据集上具有较好的分类与鉴别能力。
Aiming at the high similarity and randomness of handwriting, this paper proposes a handwriting identification which uses geometric moments to quantitatively represent the shape features of characters. The algorithm built a text-dependent word bag model and extracted eight invariant features of geometric moments. In the classification decision making level, we adopted the method of factor separation and feature fusion to realize text independent handwriting classification. The Top-1 discrimination rate of this algorithm in IAM data set reached 96.7%, which achieved the top four results in similar studies. The experimental results indicate that the handwriting identification based on word bag model and moment features has better classification and identification ability on Uyghur and English data sets.
作者
李新德
阿依夏木·力提甫
杨天
熊闻心
Li Xinde;Ayixiamu Litifu;Yang Tian;Xiong Wenxin(State Grid Hubei Information&Telecommunication Company Limited,Wuhan 430077,Hubei,China;School of Electronic Information,Wuhan University,Wuhan 430072,Hubei,China;School of Physics and Electronic Engineering,Xinjiang Normal University,Urumqi 838054,Xinjiang,China)
出处
《计算机应用与软件》
北大核心
2022年第7期154-158,180,共6页
Computer Applications and Software
基金
国家自然科学基金项目(61471272)
国网湖北省电力有限公司科技项目(52153318004G)
新疆维吾尔自治区高校科研计划自然科学青年项目(XJUDU2019Y032)。
关键词
笔迹鉴别
词袋文本
独立特征
融合矩特征
Handwriting identification
Word bag
Text-independent
Feature fusion
Moment feature
