OCR Quality Evaluation Using Foreground Pixel Coverage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing OCR systems lack unbiased and interpretable performance evaluations, leading to biased quality assessments and suboptimal performance in downstream processes due to reliance on external models and biased metrics.
Innovation Solution
A system and method for evaluating OCR performance by identifying foreground and background pixels, overlaying bounding boxes, and calculating coverage to provide an unbiased and interpretable measure of OCR quality, enabling objective comparisons and optimizations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If OCR systems use their own built-in metrics for accuracy evaluation, then the evaluation process is simple and self-contained, but the metrics are biased and do not reflect true OCR quality
Solution Approach 1:
The patent introduces an intermediary evaluation system that uses foreground-background pixel analysis as a mediator between the OCR system and quality assessment. This intermediary layer analyzes the masked image to determine if foreground pixels are appropriately covered by detected text regions, providing an unbiased quality metric that doesn't rely on the OCR system's own potentially biased confidence scores
Solution Approach 2:
The patent replaces the traditional OCR evaluation mechanism (which relies on software-based confidence scores and external ground truth models) with a visual-mechanical approach using foreground-background pixel masking and coverage analysis. This substitution eliminates reliance on biased software metrics and external training datasets, using instead a direct visual inspection method based on pixel-level analysis
2Measurement precision
If unbiased OCR evaluation methods are implemented using foreground-background pixel analysis, then measurement precision of OCR quality is improved, but device complexity increases due to additional processing steps
Solution Approach 1:
The patent segments the image processing into distinct stages: foreground pixel identification, background pixel identification, bounding box overlay, and coverage calculation. Each stage processes specific aspects of the image independently, making the complex evaluation process more manageable and implementable through modular computational steps
Solution Approach 2:
The patent creates a masked copy of the original image where foreground pixels are preserved and background pixels are removed. This copied masked image serves as the basis for evaluation without requiring modification of the original image, enabling repeated analysis and comparison while maintaining the integrity of source data
3Measurement precision
If large seed datasets are used for training external models to evaluate OCR, then measurement precision improves, but loss of time and computational resources increases
Solution Approach 1:
The patent enables the evaluation system to be self-sufficient by using the OCR system's own output (bounding boxes and detected text) as input for evaluation. The method evaluates OCR quality by analyzing the relationship between detected text regions and foreground pixels in the masked image, eliminating the need for external ground truth datasets or additional training data
Solution Approach 2:
The patent performs preliminary actions by identifying foreground and background pixels before the actual OCR evaluation takes place. This preprocessing step creates the masked image and pixel classification framework in advance, enabling rapid evaluation without requiring time-consuming dataset processing during the evaluation phase
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A processor may receive an image and determine a number of foreground pixels in the image. The processor may obtain a result of optical character recognition (OCR) processing performed on the image. The processor may identify at least one bounding box surrounding at least one portion of text in the result and overlay the at least one bounding box on the image to form a masked image. The processor may determine a number of foreground pixels in the masked image and a decrease in the number of foreground pixels in the masked image relative to the number of foreground pixels in the image. Based on the decrease, the processor may modify an aspect of the OCR processing for subsequent image processing.