Grid-Less Table Parsing Using AI Grid Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing table processing technologies, particularly for grid-less tables in fixed-layout documents, struggle with accurate parsing and data extraction, leading to manual intervention and loss of data hierarchy.
Innovation Solution
A method involving a deep learning model (DLM) for bounding box detection, followed by dilation and a conditional generative adversarial network (cGAN) for grid generation, enhances the processing of grid-less tables, maintaining data hierarchy and enabling automatic extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If existing parsing tools are used for grid-less tables, then manual selection and extraction is required, but this increases time consumption and loses data hierarchy
Solution Approach 1:
The system enables automatic table detection and extraction without manual intervention. The deep learning model autonomously identifies table regions, detects grid structures, and extracts data hierarchy, allowing the system to serve itself rather than requiring analyst intervention for each step.
Solution Approach 2:
The patent replaces manual mechanical selection processes with automated deep learning-based detection. The neural network models substitute human analysts' visual inspection and manual selection actions, enabling automated identification and extraction of tables with preserved data hierarchy.
2Measurement precision
If existing parsing tools process tables, then basic table extraction is achieved, but accuracy deteriorates for grid-less tables without proper grid structure detection
Solution Approach 1:
The patent introduces an intermediary deep learning-based detection system between the input image and the final table extraction. This intermediary layer specifically detects grid structures and table regions, mediating the complex task of accurate table parsing by first identifying the structural framework before extracting data.
Solution Approach 2:
The system changes the detection parameters and approach by using deep learning models trained specifically for table structure recognition. Instead of traditional image processing parameters, the system uses neural network parameters that can adapt to various table layouts and grid structures, improving accuracy across different document types.
3Productivity
If manual table extraction is performed, then data hierarchy can be maintained, but productivity decreases due to manual effort requirements
Solution Approach 1:
The automated system performs all table extraction operations autonomously, from detection to data extraction, eliminating the need for manual operations. The system serves itself by automatically identifying tables, detecting grids, and extracting data while maintaining hierarchy, thereby increasing productivity without sacrificing ease of operation.
Solution Approach 2:
The patent replaces manual mechanical operations with automated computational processes. The deep learning models substitute human cognitive and manual tasks, enabling high-speed automated table extraction that maintains data hierarchy while dramatically improving productivity compared to manual methods.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In a first aspect, the invention pertains to a computer-implemented method for improved grid-less table processing. In a second and third aspect, the invention pertains to a computer system and a computer program product for improved grid-less table processing. In a fourth aspect, the invention pertains use of any of the method, system or product for document parsing.