Document Splitting Using VLM Page-Set Boundary Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional document splitting methods, particularly for scanned multipage documents, face challenges in accurately determining document boundaries due to variations in layout, presence of non-textual elements, and inconsistencies in text quality, leading to inaccurate and unreliable results.
Innovation Solution
A method utilizing a multimodal vision-language model (VLM) engine that analyzes overlapping sets of image files based on visual features, such as font-style, formatting, and layout, in conjunction with text content, to determine document boundaries, and provides summaries and reasons for the determination, enhancing accuracy and reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional text-based methods are used for document splitting, then the process is simple to implement, but the accuracy and reliability of determining document boundaries deteriorates due to variations in layout, non-textual elements, and text quality inconsistencies
Solution Approach 1:
The patent combines multiple analysis modalities (visual features including layout, formatting, font-style, and text content) into a unified document splitting system. By merging these different feature types together and analyzing them in combination, the system achieves more accurate document boundary determination than text-based methods alone, while the integrated approach manages the complexity through a coordinated multi-feature analysis framework
Solution Approach 2:
The system employs a multi-functional analysis approach that evaluates both visual features (layout, formatting, font-style) and text content simultaneously. This universal method can handle various document types and boundary scenarios using the same combined feature set, improving accuracy across different document formats while maintaining a consistent processing framework
2Productivity
If manual document splitting is performed, then accuracy can be maintained through human judgment, but the time consumption and labor requirements increase significantly
Solution Approach 1:
The patent replaces manual human judgment (mechanical process) with an automated computational system that analyzes visual features and text content. This substitution enables rapid processing of documents while maintaining high accuracy through multi-feature analysis, dramatically reducing the time required compared to manual splitting while preserving quality through systematic feature evaluation
3Reliability
If computational document splitting is performed using simple methods, then the processing speed is fast, but the reliability of determining when one constituent document ends and the next begins deteriorates
Solution Approach 1:
The system merges visual feature analysis (layout, formatting, font-style) with text content analysis to improve reliability. By combining these multiple indicators, the system achieves more reliable document boundary determination than simple text-based methods, while the integrated analysis framework manages complexity through coordinated processing of all features together
Data Source
AI summary
The present disclosure provides methods and apparatuses for document splitting on a combined document file that includes, for each page of the combined document file, generating an image file that includes the contents of the page, generating a sequence of overlapping sets of image files, each set including N image files, inputting each set of N image files to a multimodal vision-language model (VLM) engine to determine whether the N image files in the set belong to a same document or to different documents based on the visual features of the image files included in the set, for each set, receiving an output that indicates whether the N image files belong to the same document or to different documents, and generating an index that correlates each page of the combined document file to a corresponding one of the one or more constituent documents based on the outputs.