Raster Image Text Detection via Shape Chain Straightening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Optical Character Recognition (OCR) methods are ineffective in detecting and processing short, curved, or text on busy backgrounds, leading to high error rates and inability to recognize text of different colors or unusual fonts embedded within other objects.
Innovation Solution
A method involving raster-to-vector conversion, shape pair detection, chain formation, curvature analysis, and straightening of text candidates, followed by classification using an automatic text classifier to identify and classify text in raster images, including those with varied colors and fonts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional OCR preprocessing is used, then rectangular text lines of sufficient length can be recognized with very low error rates, but text which is short, curved, or on a busy background cannot be detected effectively
Solution Approach 1:
The patent segments text detection into multiple processing stages: candidate block detection, chain formation by connecting adjacent blocks, curvature analysis, and straightening transformation. This segmentation allows the system to handle different text types (short, curved, various orientations) separately and systematically, improving adaptability while maintaining precision through specialized processing for each text characteristic
Solution Approach 2:
The patent transforms curved text chains into straight lines through geometric transformation, effectively changing the dimensional orientation of the text data. By converting curved paths into linear representations, the system can apply standard OCR recognition techniques to text that would otherwise be undetectable, thereby expanding detection capability without sacrificing accuracy
2Adaptability or versatility
If OCR preprocessing is applied to images with text of different colors or unusual fonts, then the system cannot detect these text types, but applying no preprocessing maintains the ability to process standard text
Solution Approach 1:
The patent applies local quality analysis by examining the geometric and topological properties of individual candidate blocks and chains rather than applying uniform preprocessing to the entire image. By analyzing local characteristics such as block adjacency, chain curvature, and shape geometry, the system can reliably detect text of different colors and fonts while maintaining recognition reliability through property-based classification
3Adaptability or versatility
If the system processes text on busy backgrounds, then detection capability is lost, but if the system avoids such processing, then standard text recognition remains reliable
Solution Approach 1:
The patent extracts text candidate blocks from busy backgrounds by detecting geometric properties and spatial relationships that distinguish text from background elements. By extracting and isolating candidate blocks based on their structural characteristics, then forming chains and analyzing their geometric properties, the system can detect text on complex backgrounds while maintaining detection precision through property-based filtering
Data Source
AI summary
Systems, methods, and applications for detection text in a raster image include converting a raster image into a vector representation of the image, identifying pairs of shapes of similar size and within a predefined distance of one another, forming shape graphs from the identified shape pairs, identifying chains of shapes from the formed shape graphs, determining characteristic chain lines associated with the identified chains of shapes, straightening the identified chains of shapes into a straight line based on the corresponding chain lines associated with the respective identified chains of shapes, and classifying the straightened identified chains as text or non-text using an automatic text classifier.