Document Image Segmentation via Aligned Text Masking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document image analysis systems struggle to accurately segment and classify connected components containing mixed types of markings, such as machine printed and handwritten text, especially when they touch graphics or annotations, leading to incorrect parsing and classification.
Innovation Solution
A system and method that involves generating an aligned text mask to break connected components into smaller fragments by detecting and segregating them based on their likelihood of belonging to aligned text, using a two-stage classifier that considers both fragment properties and neighborhood context to classify markings into specific categories like handwritten, machine printed, or noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If connected components are segmented using traditional methods, then processing speed is maintained, but segmentation accuracy deteriorates when mixed marking types are present
Solution Approach 1:
The patent divides the connected component segmentation process into multiple stages: initial connected component formation, text line detection, and fragment separation. This multi-stage segmentation approach improves accuracy by handling different marking types at appropriate levels of detail without requiring a single complex segmentation algorithm.
Solution Approach 2:
The patent introduces an aligned text mask as an intermediary element that guides the fragmentation process. This mask, generated from detected text lines, serves as a mediator to identify regions likely to contain aligned text, enabling more accurate separation of mixed markings without directly analyzing complex pixel patterns.
2Measurement precision
If a single classifier is used for marking classification, then system complexity is reduced, but classification accuracy deteriorates for mixed marking types
Solution Approach 1:
The classification system is divided into two independent classifiers: a text line detector that identifies aligned text regions, and a marking type classifier that categorizes fragments. This segmentation of classification functions allows each classifier to specialize in specific tasks, improving overall accuracy without requiring a single overly complex classifier.
Solution Approach 2:
The patent applies different classification strategies to different regions: the aligned text mask guides one classification approach for text regions, while a separate classification process handles non-text regions. This local differentiation allows each classifier to be optimized for its specific domain, improving accuracy for mixed marking types.
3Reliability
If connected components are not broken into fragments, then processing speed is maintained, but parsing accuracy deteriorates when markings touch
Solution Approach 1:
The patent performs preliminary text line detection and generates an aligned text mask before the final fragmentation step. This preliminary action identifies regions likely to contain aligned text, allowing the fragmentation process to focus computational effort only where needed, thus maintaining parsing accuracy while minimizing the impact on processing speed.
Solution Approach 2:
The patent applies fragmentation selectively rather than universally: connected components are broken into fragments primarily in regions identified by the aligned text mask, while other regions may retain their original connected component structure. This partial application of fragmentation maintains accuracy for critical text regions while preserving processing efficiency.
Data Source
AI summary
Methods and systems of the present embodiment provide segmenting of connected components of markings found in document images. Segmenting includes detecting aligned text. From this detected material an aligned text mask is generated and used in processing of the images. The processing includes breaking connected components in the document images into smaller pieces or fragments by detecting and segregating the connected components and fragments thereof likely to belong to aligned text.