Feature-Based Text Line Detection Through Region Linking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional text line detection methods struggle to accurately and efficiently detect text lines in natural scene images due to the high diversity in text appearance and surrounding backgrounds, with top-down approaches being unreliable for skewed or curved text lines and bottom-up approaches failing to robustly group irregularly organized text elements.
Innovation Solution
A method involving text region detection, link relationship determination, and text line region identification using feature representations extracted from pairs of text regions, employing machine learning models to improve the accuracy and efficiency of text line detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If top-down approaches are used for text line detection, then detection speed is improved, but accuracy deteriorates for skewed or curved text lines
Solution Approach 1:
The patent segments text detection into two phases: first detecting individual text elements using top-down approach for speed, then grouping them into text lines using bottom-up approach for accuracy. This segmentation allows each method to operate in its optimal performance zone.
Solution Approach 2:
The patent merges top-down and bottom-up approaches into a unified detection system. The top-down detector identifies candidate text elements quickly, while the bottom-up grouper accurately assembles them into text lines, combining the speed advantage of top-down with the accuracy advantage of bottom-up.
2Measurement precision
If bottom-up approaches are used for text line detection, then accuracy is improved for irregular text elements, but detection efficiency deteriorates
Solution Approach 1:
The patent performs preliminary text element detection using top-down approach before applying bottom-up grouping. This preliminary action identifies candidate text elements that need to be grouped, reducing the search space and computational burden of the subsequent bottom-up grouping process.
Solution Approach 2:
The patent segments the detection process into element detection and line grouping stages, allowing the computationally intensive bottom-up approach to operate only on detected text elements rather than the entire image, thus improving efficiency.
3Device complexity
If conventional methods are used to handle diverse text appearances and backgrounds, then system complexity is reduced, but detection reliability deteriorates
Solution Approach 1:
The patent changes detection parameters dynamically based on text characteristics. Different detectors and groupers are selected or configured based on text orientation, size, and background complexity, allowing the system to adapt to diverse text appearances without requiring a completely complex system architecture.
Solution Approach 2:
The patent creates a universal text detection system that handles multiple text types (horizontal, skewed, curved, various sizes) through a unified framework combining multiple detectors and groupers, making the system multi-functional while maintaining reasonable complexity.
Data Source
AI summary
Implementations of the present disclosure provide a solution for text line detection. In this solution, a first text region comprising a first portion of at least a first text element and a second text region comprising a second portion of at least a second text element are determined from an image. A first feature representation is extracted from the first text region and a second feature representation is extracted from the second text region. The first and second feature representations comprise at least one of an image feature representation or a semantic feature representation of the image. A link relationship between the first and second text regions can then be determined based at least in part on the first and second feature representations. The link relationship can indicate whether the first and second portions of the first and second text elements are located in a same text line. In this way, by detecting text regions and determining the link relationship thereof based on their feature representations, the accuracy and efficiency for detecting text lines in various images can be improved