Document Image to Tagged Text Conversion via Attribute Rules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional format conversion systems cannot effectively convert document image data into semantically-meaningful tagged text data, limiting accessibility for the visually impaired and failing to accurately process document attributes like font type, size, and position.
Innovation Solution
A system that uses OCR output to generate intermediate document data, which is then processed using attribute-dependent rules and templates to convert the data into a markup language format like XML, allowing for the identification of semantically meaningful sections such as titles, authors, and footnotes, and enabling flexible data processing through a programming language like PERL.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional OCR systems are used to convert document images to text, then text extraction is achieved, but the output lacks semantic meaning and document structure information
Solution Approach 1:
The patent introduces an intermediary processing stage between OCR and final output that analyzes character attributes and applies formatting rules to reconstruct document structure. This intermediary layer adds semantic meaning without requiring a completely new system, resolving the contradiction by building upon existing OCR capabilities.
Solution Approach 2:
The conversion process is segmented into distinct stages: OCR text extraction, attribute analysis of extracted text, application of formatting rules based on attributes, and generation of structured output. This segmentation allows each stage to focus on specific tasks, improving information retention while managing system complexity through modular design.
2Loss of information
If detailed attribute analysis is performed on OCR output to identify document structure, then semantic meaning is improved, but processing time and computational resources increase
Solution Approach 1:
The system changes parameters by analyzing specific attributes of OCR output (character spacing, font variations, positioning) rather than processing entire documents uniformly. This targeted parameter analysis recovers document structure while minimizing processing time by focusing computational resources on discriminative features.
Solution Approach 2:
Different processing strategies are applied to different portions of the document based on local characteristics. Sections with clear structural attributes receive detailed analysis, while uniform sections are processed more quickly. This local quality approach optimizes the balance between structure recovery and processing efficiency.
3Loss of information
If conventional format conversion systems process document data, then basic text conversion is achieved, but image data and document attributes are not preserved
Solution Approach 1:
The system performs preliminary analysis of OCR output attributes before applying conversion rules. By pre-identifying character attributes and their relationships, the system preserves document attributes and image-derived information while maintaining conversion efficiency through prepared processing pipelines.
Solution Approach 2:
The conversion system is designed to handle multiple document types and attributes through a universal framework that processes various attribute combinations. This multi-functionality allows preservation of diverse document characteristics without requiring separate processing paths, maintaining productivity across different document formats.
Data Source
AI summary
Converting intermediate document data representing document text derived from data in an image data format into a semantically-meaningful tagged text data format may be provided. Intermediate document data derived from document image data may be imputed. The intermediate document data may comprise character data corresponding to characters in the document and attribute data corresponding to one or more attributes of characters in the document. The intermediate document data may then be processed according to attribute-dependent rules. Tagged text input data may be generated comprising tagged section of the document text. The tags may define semantically meaningful portions of the text determined according to the attribute data.