摘要
文字的识别过程需要提取文字的骨架信息,但在文字骨架提取的过程中往往会产生毛刺,毛刺的出现会影响文字识别性能。以往的毛刺删除方法仅考虑了毛刺长度阈值,这可能会误删骨架本身,并且仅考虑了两种节点的情况,并不具备普适性。针对文字骨架毛刺的特点,引入节点冗余去除模板,并在毛刺长度阈值筛选的基础上,根据主干、分支的方向连续特性再一次界定毛刺,最终删除文字骨架毛刺。实验表明,方向约束法能够很好地保留文字本身骨架,准确率达到97.1%,相比阈值法准确率提高了4.73%,具有一定理论、实用价值。
The character recognition process needs to extract skeleton information of character,but in the process of character skeleton extraction,the appearance of the burr will affect the performance of the character recognition.The previous burr deletion method only considers threshold of burr length,which may accidentally delete the skeleton itself,and only considers two kinds of nodes,so it is not universal.According to the characteristics of character skeleton burr,a node redundancy removal template was introduced,and on the basis of threshold selection of burr length,the burr was defined again according to the direction continuous of trunk and branch,and finally character skeleton burr was deleted.The experimental results show that direction constraint method can retain itself well,and the accuracy rate is increased by 4.73%,reaching 97.1%compared with the threshold method,which has certain theoretical and practicalvalues.
作者
范静
颜廷秦
崔鸣
薛知言
FAN Jing;YAN Tingqin;CUI Ming;XUE Zhiyan(School of Electronic Information Engineering,Suzhou Vocational University,Suzhou 215104,China)
出处
《苏州市职业大学学报》
2021年第3期18-23,共6页
Journal of Suzhou Vocational University
基金
苏州市职业大学校级科研平台智能计算与知识学习研究中心(201804000053)。
关键词
文字骨架
毛刺
长度阈值
方向约束
链码跟踪
character skeleton
burr
length threshold
direction constraint
chain code tracking
