基于极坐标转换的中文印章文字识别  

Chinese seal text recognition based on polar coordinate conversion

作　　者：戴俊峰 杨天 熊闻心 DAI Jun-feng;YANG Tian;XIONG Wen-xin(State Grid Hubei Information and Telecommunication Limited Company,Wuhan 430077,China;School of Electronic Information,Wuhan University,Wuhan 430072,China)

机构地区：[1]国网湖北省电力有限公司信息通信公司,湖北武汉430077 [2]武汉大学电子信息学院,湖北武汉430072

出　　处：《计算机工程与设计》2021年第11期3174-3180,共7页Computer Engineering and Design

基　　金：国网湖北省电力有限公司科技基金项目(52153318004G);国家自然科学基金项目(61471272)。

摘　　要：为能够处理文档中印章元素的关键信息,促进办公智能化,提出一种基于极坐标转换的印章文字识别方法。根据印章元素通常成圆环状排列的特点,对中文印章图像进行极坐标展开,克服印章文字方向不统一的问题,利用CTPN+CRNN网络进行文字的检测与识别,对CTPN网络中的文本构造算法进行改进,实现对印章内容快速准确的识别。用该算法对自制的中文印章数据集进行实验,印章内容的文字识别召回率可以达到90.4%,表明该算法可以有效检测识别印章内容,对文档的分类与鉴别研究具有重要的意义。To process the key information of seal elements in documents and promote the intelligence of office,a method of seal character recognition based on polar coordinate transformation was proposed.According to the characteristics of seal elements that were usually arranged in circles,the polar coordinates of Chinese seal images were developed to overcome the problem of inconsistent seal character direction.CTPN+CRNN network was used for character detection and recognition,and the text construction algorithm in CTPN network was improved to realize the fast and accurate recognition of seal content.The algorithm was used on the self-made Chinese seal data set,the recall rate of the seal content can reach 90.4%,indicating that this algorithm can effectively detect and identify the seal content,which is of great significance to the classification and identification research of documents.

关 键 词：印章识别 极坐标转换 文档鉴别 文字分割 文字检测 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]