Document Image Graphic Removal via Heuristic Text Analysis and Recovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for removing graphics from document images often fail to effectively separate text from graphics, leading to incomplete text extraction for OCR and other analyses, as they either leave behind significant graphic components or inadvertently remove text due to stringent criteria.
Innovation Solution
A method that employs a two-stage process: first, an aggressive graphic removal stage using heuristic text analysis to distinguish and remove graphic components, and second, a text recovery stage that re-examines the image to recover mistakenly removed text components by expanding bounding boxes around remaining text components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If aggressive graphic removal criteria are used to remove all graphic components, then graphic removal completeness is improved, but text loss increases
Solution Approach 1:
The patent segments the graphic removal process into two distinct stages: an aggressive removal phase that prioritizes complete graphic elimination, followed by a recovery phase that selectively restores text components. This segmentation allows each stage to optimize for its specific goal without compromising the other.
Solution Approach 2:
The patent deliberately discards text components during the aggressive removal phase, then recovers them in the second phase by analyzing spatial relationships with remaining text. This temporary discarding and subsequent recovery resolves the contradiction by allowing complete graphic removal followed by text restoration.
2Loss of information
If stringent removal criteria are used to ensure text preservation, then text loss is reduced, but graphic removal completeness deteriorates
Solution Approach 1:
The patent performs preliminary aggressive graphic removal without concern for text loss, then applies a recovery action in the second stage. This preliminary action allows the system to achieve complete graphic removal first, then correct any text loss subsequently.
Solution Approach 2:
The patent uses feedback from the spatial relationships between components to guide the recovery process. By analyzing which removed components are adjacent to remaining text components, the system intelligently restores only the text that was mistakenly removed, achieving both complete graphic removal and text preservation.
3Productivity
If simple size-based criteria are used for component classification, then processing speed is improved, but classification accuracy deteriorates
Solution Approach 1:
The patent employs dynamic classification criteria that adapt based on the component's spatial context. Rather than using fixed size thresholds, the system adjusts classification decisions based on the component's relationship to other components, particularly its proximity to text regions, thereby improving accuracy without significantly impacting speed.
Data Source
AI summary
A graphic removal process for document images involves two stages: First, removal of graphics in the document image based on heuristic text analyses; and second, text recovery to recover some text that is accidentally removed during the first stage. The first stage uses a relatively aggressive strategy to ensure that all graphics components are removed, which also temporarily leads to the removal of some text; the lost text will then be recovered using the text recovery technique. The heuristic text analyses utilize the geometric properties of text characters and consider the properties of text characters in relation to their neighbors. The text recovery technique starts from the text that remain after the first stage, and recovers any connected component that is at least partially located within a pre-defined neighboring area around any of the text components in the intermediate document image.