Arbitrary Shape Text Region Detection via Geometry Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional text detection methods struggle to produce compact bounding boxes for non-rectangular texts, such as arc-shaped or circular texts, leading to inaccurate text recognition and processing.
Innovation Solution
The proposed method includes an arbitrary shape text detecting module that adjusts the initial text region based on text geometry information, such as centerlines and border distances, to generate a more compact and accurate text bounding box, which can be integrated with universal text detection modules to improve text detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional text detection methods use standard bounding boxes to cover text regions, then all text regions can be detected, but the bounding boxes are not compact for non-rectangular texts leading to reduced recognition accuracy
Solution Approach 1:
The patent segments the text detection process into two stages: first detecting the text region with a standard bounding box, then further segmenting and adjusting the bounding box based on the actual text shape characteristics. This segmentation allows the system to handle both rectangular and non-rectangular texts effectively by applying different refinement strategies.
Solution Approach 2:
The patent introduces geometric transformation by converting the standard rectangular bounding box into a polygonal bounding box that adapts to the text shape. This dimensional transformation from fixed rectangle to flexible polygon allows the bounding box to conform to arc-shaped, circular, and other non-rectangular text layouts, significantly improving compactness and recognition accuracy.
2Measurement precision
If the text region is adjusted to be more compact for non-rectangular texts, then recognition accuracy improves, but the detection process becomes more complex
Solution Approach 1:
The patent performs preliminary detection of text regions using standard bounding boxes before applying shape-based adjustments. This preliminary action identifies candidate text regions that require further processing, allowing the system to apply complex geometric transformations only where necessary rather than to all detected regions, thus managing computational complexity.
Solution Approach 2:
The patent introduces an intermediary geometric transformation module that acts as a bridge between the simple standard bounding box detection and the final compact polygonal bounding box. This intermediary layer processes the text region information and applies appropriate geometric transformations based on text shape characteristics, making the overall system more manageable and modular.
3Device complexity
If standard bounding boxes are used for all text regions, then the detection process is simple, but non-rectangular texts such as arc-shaped or circular texts are not accurately covered
Solution Approach 1:
The patent applies local quality by differentiating the treatment of different text regions based on their shape characteristics. Standard rectangular bounding boxes are applied to regular text regions, while polygonal bounding boxes with geometric transformations are applied to non-rectangular text regions such as arc-shaped or circular texts. This localized approach ensures high coverage accuracy for each text type without unnecessarily complicating the overall detection process.
Data Source
AI summary
A method and apparatus for detecting text regions in an image, a device, and a medium are provided. The method may include: detecting, based on feature representation of an image, a first text region in the image, where the first text region covers a text in the image, a region occupied by the text being of a certain shape; determining, based on a feature block of the first text region, text geometry information associated with the text, where the text geometry information includes a text centerline of the text and distance information of the centerline from the upper and lower borders of the text; and adjusting, based on the text geometry information associated with the text, the first text region to a second text region, where the second text region also covers the text and is smaller than the first text region.