Table Recognition Using Separator-Aware Cell-Text Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing table recognition methods fail to accurately match the location information of table cells with their corresponding texts due to insufficiently accurate location information, leading to incorrect table extraction and degraded user experience.
Innovation Solution
A comprehensive table recognition method that involves obtaining and processing a target image to separately identify cells and separators, performing text recognition, and calculating location information of both cells and separators, thereby ensuring accurate matching through a multi-task framework that considers the impact of separators between cells.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If only the location information of cells output by the model is considered, then the matching process is simple, but the matching accuracy deteriorates when location information is insufficiently accurate
Solution Approach 1:
The patent introduces separator location information as an intermediary element to bridge the matching between cell location information and text location information. By obtaining location information of separators between cells and using it to calculate refined cell location information, the system improves matching accuracy without directly complicating the core matching process. The separator acts as a mediator that provides additional spatial context for more accurate alignment.
2Measurement precision
If comprehensive processing is performed on multiple location information sources, then matching accuracy is improved, but the processing complexity increases
Solution Approach 1:
The patent performs preliminary processing to obtain separator location information before the actual cell-text matching process. By pre-calculating the location information of separators and using it to derive refined cell location information in advance, the system prepares more accurate spatial data for matching. This preliminary action reduces the complexity during the actual matching phase while maintaining high accuracy.
Solution Approach 2:
The patent segments the location information processing into distinct components: cell location information, separator location information, and text location information. By separating these elements and processing them independently before integration, the system manages complexity more effectively. Each component is processed and validated separately, then combined to achieve accurate matching without overwhelming processing complexity.
3Productivity
If the location information of cells is insufficiently accurate, then the processing speed is maintained, but the table recognition correctness deteriorates
Solution Approach 1:
The patent uses separator location information as an intermediary to improve the accuracy of cell location information without significantly impacting processing speed. The separator-based refinement provides a computationally efficient way to enhance location accuracy by leveraging the spatial relationship between separators and cells, rather than requiring complex re-processing of all location data.
Data Source
AI summary
This application discloses a table recognition method. The method includes: a target image including a to-be-recognized table can be first obtained. The target image can be processed using a target model, to separately obtain a plurality of cells of the table, first location information of the plurality of cells, and second location information of a plurality of separators between the plurality of cells. Text recognition can be further performed on the target image, to obtain a plurality of texts of the table and third location information of the plurality of texts. Then, fourth location information of the plurality of cells can be determined based on the second location information of the plurality of separators. Finally, the location information can be comprehensively considered, to complete matching between the plurality of cells and the plurality of texts, so as to obtain the table.