OCR Text Region Quality Feedback via Multi-Zoom Capture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing OCR methods fail to accurately recognize text in natural images or video frames due to variations in lighting, color, tilt, focus, and font, resulting in high false positives and negatives, making them impractical for real-world applications.
Innovation Solution
An electronic device and method that captures multiple images of a scene at various zoom levels, applies a 'text-presence' test to identify potential text regions, and checks image quality at a region level, providing feedback to improve image quality before performing OCR, allowing for accurate text recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If prior art OCR methods are applied to natural images, then text recognition can be performed, but high false positives and negatives occur due to variations in lighting, color, tilt, focus, and font
Solution Approach 1:
The system performs preliminary quality assessment of text regions before applying OCR, checking attributes such as resolution, lighting conditions, tilt angle, and text size. This preliminary action identifies unsuitable regions that would produce false positives or negatives, allowing the system to request re-captures with improved quality parameters before recognition attempts
Solution Approach 2:
The system implements a feedback loop where quality assessment results are fed back to guide subsequent image capture or processing. When text regions fail quality checks, the system provides feedback to request re-captures with specific improvements (e.g., better lighting, reduced tilt, closer distance), continuously improving recognition reliability
2Reliability
If multiple images at various zoom levels are captured and analyzed, then text recognition accuracy improves, but device complexity and processing time increase
Solution Approach 1:
The system segments the image processing task by analyzing text regions from multiple images at different zoom levels separately. Each image is processed independently to identify text regions, and results are aggregated. This segmentation allows the system to leverage multiple views while maintaining manageable processing complexity through modular region analysis
Solution Approach 2:
The system applies partial action by selectively processing only those text regions that meet quality criteria from the multiple captured images. Rather than processing all regions from all images equally, it focuses computational resources on high-quality regions that are most likely to yield accurate recognition, reducing unnecessary processing overhead
3Reliability
If region-level image quality checking is performed before OCR, then false positives and negatives are reduced, but processing time increases due to additional analysis steps
Solution Approach 1:
The system performs preliminary quality checking of text regions using efficient attribute assessments (resolution, lighting, tilt, size) before committing to time-consuming OCR processing. This preliminary filtration prevents wasted processing time on low-quality regions that would fail OCR anyway, reducing overall processing time while maintaining high accuracy
Solution Approach 2:
The system replaces direct OCR processing with a preliminary quality assessment mechanism that uses simpler, faster attribute checks. This substitution filters out unsuitable regions before they undergo complex OCR processing, trading a small amount of assessment time for significant savings in avoided OCR operations on poor-quality regions
Data Source
AI summary
An electronic device and method capture multiple images of a scene of real world at a several zoom levels, the scene of real world containing text of one or more sizes. Then the electronic device and method extract from each of the multiple images, one or more text regions, followed by analyzing an attribute that is relevant to OCR in one or more versions of a first text region as extracted from one or more of the multiple images. When an attribute has a value that meets a limit of optical character recognition (OCR) in a version of the first text region, the version of the first text region is provided as input to OCR.