Text Detection Using Image Region Stability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional text detection methods fail to reliably detect text with colors similar to the background, and are computationally demanding, making them unsuitable for portable devices.
Innovation Solution
The method involves blob-based text extraction, where stable and curvilinear image regions are identified and processed to generate geometric figures, allowing for projection profile analysis to estimate skew and tilt, reducing computational load and enabling text recognition without requiring distinct text and background colors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If color segmentation methods are used for text detection, then text detection accuracy is improved for text with similar colors to background, but computational demand increases making it unsuitable for portable devices
Solution Approach 1:
The patent segments the text detection process into multiple stages: initial binarization to separate text from background, followed by projection profile analysis to identify text lines, and finally character recognition. This staged segmentation allows the system to process only relevant regions in detail, reducing overall computational demand while maintaining accuracy for text with similar colors to background.
Solution Approach 2:
The patent extracts text regions from the image using projection profile analysis after binarization, isolating only the text-containing areas for further processing. This extraction approach avoids applying computationally intensive color segmentation to the entire image, reducing computational demand while maintaining detection accuracy for text with similar colors to background.
2Measurement precision
If projection profile analysis is performed on binary images, then skew and tilt estimation is improved, but computational demand increases making it unsuitable for portable devices
Solution Approach 1:
The patent performs binarization as a preliminary action before projection profile analysis, creating a simplified binary image that enhances the visibility of text structures. This preliminary processing reduces the complexity of subsequent projection profile analysis, allowing accurate skew and tilt estimation with lower computational demand suitable for portable devices.
Data Source
AI summary
A method includes receiving an indication of a set of image regions identified in image data. The method further includes, selecting image regions from the set of image regions for text extraction at least partially based on image region stability.