Chart document panel analysis and understanding method

Abstract

The invention discloses a chart document panel analysis and understanding method, which comprises the following steps of: collecting an initial chart document, and analyzing the initial chart document to obtain a chart analysis data set; constructing a key point segmentation model, and obtaining the position coordinates of the scale points based on the chart analysis data set and the key point segmentation model; constructing a multi-target detection model, carrying out legend position detection based on the multi-target detection model, and obtaining legend positions; and analyzing the position coordinate of the scale point and the legend position based on a matching rule of the scale point and the scale value of the rectangular expansion and a matching rule of the legend and the label of the maximum intersection-to-union ratio to obtain an analysis result of the chart document panel. According to the method provided by the invention, the problem of panel understanding of the chart document is efficiently and accurately solved, and the method plays an important positive role in realizing automatic data extraction and content understanding of the chart document by a machine.

Description

technical field [0001] The present invention relates to the technical field of image precise positioning and element matching, in particular to a method for analyzing and understanding chart document panels. Background technique [0002] As a special type of document containing rich semantic information, graphs can display data in a more intuitive format, and are usually used to summarize experimental results or conclusions. The general document analysis and recognition (DAR) task has been studied for decades, but chart document analysis and recognition (CHART-DAR) is still an urgent problem to be solved. Because charts have their own characteristics, such as diversity, format uncertainty, and semantics, the CHART-DAR task is very different and difficult from other document analysis tasks. The analysis and understanding of chart documents involves multiple subtasks. One of the basic and important issues is how to understand the basic physical structure of charts, including ...

AI Technical Summary

Problems solved by technology

In recent years, with the development of deep neural networks, related deep learning-based methods have begun to focus on the problem of chart document analysis, but existing work tends to solve the problem of any specific type of chart, such as scatter plots, bar charts or pie charts, and the proposed method is not suitable for some more complex charts

Images