摘要
为了快速、准确地识别出文档图片中存在的表格,为表格信息提取提供表格图像数据,为表格内容的语义分割打下基础。本文首先使用OpenCV图像处理工具对包含表格的文档图片进行预处理,再采用Labelme标注工具对图片中的表格位置进行标注;其次,把图片数据集按照4:1比例分为表格识别模型训练数据集和模型验证数据集;最后,借助Tensorflow深度学习工具,采用Faster-RCNN目标检测框架对表格识别模型进行训练,并用验证数据集对训练好的模型进行验证实验。实验结果表明,基于Faster-RCNN算法的表格检测模型系统平均每张图片的处理时间为1.31s,识别准确率达到92.4%。说明Faster-RCNN目标检测算法能准确且快速地检测出文档图像中存在的表格。
In order to identify the table in the document image more quickly and accurately,make a foundation for the table information extraction.This paper using the OpenCV image processing tools to preprocess the document images containing tables,then uses the Labelme labeling tool to mark the position of the table in the picture,and the image data set is divided into training data and verification data set according to the ratio of 4:1,finally,the Tensorflow deep learning tool is used to train the table recognition model and the target detection framework of Faster-RCNN is adopted.The trained model was validated with the validation data set.The experimental results show that the processing time of the table detection algorithm system based on FasterRCNN is 1.31 s,and the recognition accuracy reaches 92.4%.It means that Faster-RCNN can accurately and quickly detect the existing table in the document.
作者
马志远
余粟
MA Zhiyuan;YU Su(School of Mechanical and Automotive Engineering,Shanghai University of Engineering Science,Shanghai 201620,China;Engineering Training Center,Shanghai University of Engineering Science,Shanghai201620,China)
出处
《智能计算机与应用》
2020年第12期24-27,31,共5页
Intelligent Computer and Applications
作者简介
马志远(1995-),男,硕士研究生,主要研究方向:机器视觉;通讯作者:余粟(1962-),女,硕士,教授,硕士生导师,主要研究方向:计算机科学,Email:suyu_sh@hotmail.com。
