Table structure identification method and device, storage medium and electronic equipment

Abstract

The invention relates to a table structure recognition method and device, a storage medium and electronic equipment, a table image corresponding to a to-be-recognized table can serve as input of a target detection model, the position feature and the element category of each table element in the to-be-recognized table are obtained, and the element category comprises text lines; character recognition is performed on the character lines through a character recognition model to obtain a text semantic feature of each character line; feature extraction is performed on the table image through the image feature extraction model to obtain an image feature map, and the image feature map is samplied to obtain a target image feature corresponding to each table element; according to the position features, the text semantic features and the target image features, target relation features used for representing the topological relation between every two table elements in the to-be-recognized table are obtained through a preset relation extraction model, a graph adjacency matrix used for representing the table structure of the to-be-identified table is determined by a preset classification model according to the target relation characteristics.

Description

technical field [0001] The present disclosure relates to the field of table structure recognition, and in particular, to a table structure recognition method, device, storage medium and electronic equipment. Background technique [0002] As an important subfield of OCR (Optical Character Recognition, Optical Character Recognition) technology, table structure recognition plays an indispensable role in many text recognition applications. [0003] In related technologies, graphs can be used to represent table structures, and table image features and text position features are used to obtain graphs representing table structures based on graph neural networks. However, when using text features to identify table structures, this method is only based on text positions and text characters. String length is used as a feature. The introduction of this feature does not bring substantial information gain to table structure recognition, which makes the accuracy of table structure recogni...

AI Technical Summary

Problems solved by technology

[0003] In related technologies, graphs can be used to represent table structures, and table image features and text position features are used to obtain graphs representing table structures based on graph neural networks. However, when using text features to identify table structures, this method is only based on text positions and text characters. String length is used as a feature. The introduction of this feature does not bring substantial information gain to table structure recognition, which makes the accuracy of table structure recognition insufficient. In addition, in the process of sampling image features, this method directly Sampling is performed at the center of the ROI area in the figure. The sampled image features cannot represent the image feature information of the overall table, which will also affect the accuracy of table structure recognition.

Images