Table detection method and device based on semantic segmentation, equipment and medium

Abstract

The invention relates to a table detection method and device based on semantic segmentation, computer equipment and a storage medium. According to the method, prediction can be carried out for various types of table lines, the method can adapt to table images of complex layouts, and the robustness of detection of table images of different layouts is improved. The method comprises the following steps: acquiring a table image, inputting the table image into a coding module, acquiring a first feature map from the output of the coding module, and acquiring a second feature map from the output of a backbone network in the coding module; and taking the second feature map as the input of a decoding module so as to be transmitted to a splicing layer in the decoding module by the decoding module, taking the first feature map as the input of the splicing layer so that the splicing layer fuses the first feature map and the second feature map to obtain a fused feature map, performing classification prediction on the table line based on the fused feature map, and outputting the table line, and obtaining a prediction result.

Description

technical field [0001] The present application relates to the technical field of image detection, in particular to a table detection method, device, computer equipment and storage medium based on semantic segmentation. Background technique [0002] With the development of text recognition technology, people can quickly recognize and extract text or numbers in images generated by scanning, taking pictures, etc. Among them, tables have become the focus of text recognition technology, because tables contain different structures. For example, rows and columns, levels, expression sizes, and styles are complex and diverse. The table also includes different background fills, rows and columns merged, and the text in it can also include modern, historical, handwritten, printed, etc. Different types make the table Consistent recognition is a difficult point in the field of text recognition. [0003] At present, there are two types of mainstream table detection algorithms: target dete...

AI Technical Summary

Problems solved by technology

[0003] At present, there are two types of mainstream table detection algorithms: target detection-based algorithms and segmentation-based algorithms. The diversity of preferences and other elements and the essentially free scheduling rules make the table styles tend to be diverse and complex, which leads to the fact that the existing methods are not robust and versatile when dealing with complex layout styles. ability

Images