项目申报书形式审查算法  

Formal examination algorithm for project application materials

作　　者：张彬 程健峰 Zhang Bin;Cheng Jianfeng(School of Communication,Mianyang Normal University,Mianyang 621000,China;School of Computer Science,Sichuan University,Chengdu 610065,China)

机构地区：[1]绵阳师范学院传媒学院,绵阳621000 [2]四川大学计算机学院,成都610065

出　　处：《现代计算机》2023年第10期15-19,共5页Modern Computer

摘　　要：根据项目申报书扫描件的文本较为规范的特点,采用MSER算法进行文字内容的检测,然后利用PaddleOCR进行公司名称识别,实验结果表明召回率达到91.38%,精确度为98.56%,F1值为94.83%。采用CRNN深度神经网络对印章内容检测,并根据印章文字分布的空间特征,在CNN网络中采用圆形卷积核,对比传统的方形卷积核,识别效果得到一定提升,召回率为98.90%,精确度为98.39%,F1值为98.64%。实验结果表明,文章提出的算法可以有效进行项目申报书扫描件中涉及印章匹配的形式审查,对提升机关部门的工作效率具有重要意义。In view of the relatively standardized characteristics of the project application materials,the MSER algorithm to de-tect and recognize characters,then the company name is recognized by PaddleOCR.The experimental results show that the recall rate can reach 91.38%,the accuracy is 98.56%,F-measure is 94.83%.In this paper,CRNN deep neural network is used to detect the seal content.Given the spatial characteristics of seal text distribution,circular convolution kernel is used in CNN network,the recognition effect of which is better than the traditional square convolution kernel.According to the experimental results,the recall rate can reach 98.90%,the accuracy is 98.39%,F-measure is 98.64%.The experimental results show that the algorithm proposed in this paper can effectively carry out the formal examination involving seal matching in the scanned documents of project applica-tion,and is of great significance to improve the work efficiency of the government department.

关 键 词：图像处理 印章识别 文档检测 深度神经网络 字符识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]