Dot Text Optical Character Recognition via Adaptive Dot Transformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical character recognition (OCR) methods for dot text in images face challenges with non-uniform dot spacing, deformed, skewed, and rotated characters, as well as varying contrast and backgrounds, requiring extensive pre-processing and being sensitive to dot pitches and angles.
Innovation Solution
A method that extracts dots from images based on selection criteria, transforms them into stroked characters using candidate distances along different orientations, and performs model-based character matching without prior segmentation, enabling robust recognition across varying conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional OCR methods are used for dot text, then character recognition can be performed, but the system requires extensive pre-processing and is sensitive to dot pitch variations and text orientations
Solution Approach 1:
The patent segments the dot text image into multiple candidate regions based on dot density and spatial distribution patterns. Each candidate region is independently analyzed to determine text presence and orientation, allowing the system to handle rotated and skewed text without extensive global pre-processing
Solution Approach 2:
The patent transforms the recognition problem by changing parameters from fixed dot pitch assumptions to adaptive dot density analysis. By calculating local dot densities and using these as features for candidate region identification, the system becomes robust to dot pitch variations while reducing pre-processing requirements
2Adaptability or versatility
If dot text is recognized with non-uniform dot spacing and deformed characters, then real-world applications can be handled, but measurement precision deteriorates
Solution Approach 1:
The patent employs dynamic thresholding for dot detection, where the threshold adapts to local image conditions and dot density patterns. This allows the system to maintain measurement precision across varying dot spacings and deformations by adjusting detection parameters locally rather than using fixed global thresholds
Solution Approach 2:
The patent performs preliminary dot detection and density calculation across the entire image before candidate region identification. This preliminary action establishes a foundation of dot distribution knowledge that guides subsequent precise measurements and character recognition, improving both adaptability and precision
3Productivity
If candidate regions are identified based on dot density, then text locations can be found efficiently, but false positives may occur in dense dot patterns
Solution Approach 1:
The patent implements feedback mechanisms where candidate regions are validated by checking multiple dot density patterns and spatial relationships. Regions that don't conform to expected text patterns (such as regular dot spacing ratios) are rejected, providing feedback-based filtering that reduces false positives while maintaining efficient location identification
Solution Approach 2:
The patent performs partial validation of candidate regions by checking only critical dot density features rather than complete character recognition. This partial action approach efficiently filters obvious false positives while allowing potentially valid regions to proceed to full recognition, balancing productivity and reliability
Data Source
AI summary
A method and apparatus for optical character recognition of dot text in an image are described. A plurality of dots that satisfy dot selection criteria are extracted from an image. A transformation is performed on the plurality of dots based on a first candidate distance between adjacent dots along a first orientation and on a second candidate distance between adjacent dots along a second orientation to obtain a transformed image including stroked characters. Model based character matching is performed on the transformed image to output a candidate string of characters.