Table structure reconstruction method based on deep learning

Abstract

The invention discloses a table structure reconstruction method based on deep learning, and the method comprises the following steps: obtaining a training image which displays a table; preprocessing the training image; extracting a feature map of the preprocessed image; performing learning and parameter updating through the feature map, and acquiring a table line classification and positioning model; obtaining a to-be-reconstructed image for table reconstruction, wherein the to-be-reconstructed image displays a to-be-reconstructed table; obtaining structure information of the to-be-reconstructed table according to the table line classification and positioning model; performing character recognition and image target detection on the to-be-reconstructed image to obtain table content information; matching the structure information of the to-be-reconstructed table with the table content information, and reconstructing the table. According to the method, training of the network can be completed by using less data so that the network is enabled to learn stable and accurate feature information, the accuracy of form line information extraction under the condition of low data samples can begreatly enhanced, and the algorithm has great generalization capability and great robustness.

Description

technical field [0001] The invention relates to the fields of image processing and pattern recognition, in particular, a table structure reconstruction method based on image processing and deep learning methods. Background technique [0002] Forms, as a common document format, frequently appear in people's lives, such as resumes, application forms, financial statements, etc. Table styles are changeable and have their own layout characteristics. However, when people need to use tables, they often need to create new table styles by themselves, which is very time-consuming. [0003] At present, there is no mature automation solution that can help or assist users to quickly complete table copying or editing. Most of the early table restoration or table recognition methods are traditional image processing schemes based on line regression or Hough transform as the main method, which is difficult to deal with a variety of table styles and scenarios. [0004] It can be seen that t...

AI Technical Summary

Problems solved by technology

Most of the early table restoration or table recognition methods are traditional image processing schemes based on line regression or Hough transform as the main method, which is difficult to deal with various table styles and scenarios

[0004] It can be seen that the current table reconstruction scheme has obvious problems of poor generalization ability and single applicable environment

Images