Implicit Table Extraction from Richly Formatted Documents
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods fail to accurately extract implicit tables from richly formatted documents, such as scanned files and web pages, which are composite organic typesettings of various components like titles, text chapters, and images.
Innovation Solution
A method and apparatus utilizing a deep convolutional neural network to perform table detection, through-line drawing, and table-cell merging processes, acquiring richly formatted documents, and employing preset models to obtain explicit table content by identifying and processing table tags, through-line tags, and short-line tags.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional extraction methods are used on richly formatted documents, then the extraction process is simple, but the accuracy of extracting implicit tables is poor
Solution Approach 1:
The extraction system is segmented into three independent modules: table detection model, through-line drawing model, and table-cell merging model. Each module performs a specific function in the extraction pipeline, allowing the system to handle complex implicit table structures through sequential processing stages while maintaining modularity and manageability.
Solution Approach 2:
The through-line drawing model serves as an intermediary between table detection and table-cell merging. It generates through-line tags that act as intermediate structural information, bridging the gap between detecting implicit table regions and merging table cells, thereby enabling accurate reconstruction of implicit table layouts.
2Productivity
If manual extraction methods are used, then the extraction accuracy can be maintained, but the extraction efficiency is low
Solution Approach 1:
The patent replaces manual mechanical extraction processes with automated deep learning models. The table detection model, through-line drawing model, and table-cell merging model collectively automate the entire extraction pipeline, substituting human operators with intelligent algorithms that achieve both high efficiency and high accuracy simultaneously.
3Reliability
If simple extraction algorithms are used, then the processing speed is fast, but the ability to handle implicit tables is insufficient
Solution Approach 1:
The table detection model performs preliminary action by detecting and marking implicit table regions before the through-line drawing and cell merging processes. This preliminary detection creates a foundation that guides subsequent processing steps, ensuring that implicit tables are identified and handled appropriately throughout the extraction pipeline.
Solution Approach 2:
The extraction process maintains continuity through three sequential models that work without interruption: table detection identifies implicit tables, through-line drawing reconstructs table structures, and table-cell merging finalizes the extraction. This continuous pipeline ensures implicit tables are processed reliably while minimizing idle time between operations.
Data Source
AI summary
The disclosure discloses a method, apparatus, device for table extraction based on a richly formatted document and medium. The method comprises: acquiring page content; performing a table detection process on the page content by use of a preset table detection model to obtain a list of table tags, and to obtain a first table content; performing, by use of a preset through-line drawing model, a through-line drawing process on the first table content to obtain a list of through-line tags, and to obtain a second table content; and performing, by use of a preset table-cell merging model, a table-cell merging process on the second table content to obtain a list of short-line tags, and to obtain an explicit table content.