Table Data Extraction Using Text Rectangle Feature Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated table extraction algorithms struggle to understand table content effectively, leading to challenges in higher-level information extraction, such as determining data types in columns and accurately extracting wrapped text in table cells.
Innovation Solution
The approach involves using a hierarchical process with small machine learning models to generate text-level numerical data from read-only documents, creating a feature map that can be processed by a convolutional neural network for fully automated table extraction without relying on raw image pixel data or OCR.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If image-based automated table extraction algorithms are used, then table data can be extracted automatically, but the algorithms cannot understand table content effectively for higher-level information extraction
Solution Approach 1:
The patent segments the table extraction process into distinct functional components: a table detector that identifies table locations and structures, and a data extractor that retrieves data from detected tables. This segmentation allows each component to specialize in its function, improving both automation and content understanding capabilities.
Solution Approach 2:
The patent introduces an intermediary processing layer that bridges image-based detection and content understanding. This layer processes the detected table structure to enable higher-level information extraction, acting as a mediator between the automated detection system and the semantic understanding requirements.
2Measurement precision
If template-driven or OCR approaches are used, then table extraction can be performed, but manual intervention is required and the solution is not fully automated
Solution Approach 1:
The system employs self-service mechanisms where the table detector automatically identifies table structures and the data extractor automatically retrieves data without manual template creation or OCR intervention. The system serves itself by learning from data and performing extraction tasks autonomously, achieving full end-to-end automation while maintaining precision.
3Measurement precision
If expensive large neural networks are used for table extraction, then extraction accuracy can be improved, but computational cost and complexity increase significantly
Solution Approach 1:
The patent divides the complex table extraction task into simpler sub-tasks performed by specialized components: table detection and data extraction. Each component uses a simpler neural network model optimized for its specific function, reducing overall system complexity while maintaining or improving extraction accuracy through coordinated operation of the segmented components.
Data Source
AI summary
Embodiments of the disclosed technologies provide solutions for automatically reading digital electronic documents that contain tables and correctly extracting table data, rows and columns from the documents with high accuracy and high throughput. Embodiments are capable of converting a table portion of a read-only document to a searchable, editable data record using text rectangle (TR)-level numerical data that indicates probabilities of TRs belonging to canonicals and at least one convolutional neural network (CNN) that processes the TR-level numerical data to produce table-level numerical data.