Text Recognition Using Centerline and Offset Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current end-to-end text detection and recognition technologies in natural scenarios face challenges in real-time processing and accuracy, particularly with the two-stage solution requiring separate detection and recognition steps, and the character-based method needing large backbone networks.
Innovation Solution
A method and apparatus that acquire images of a text area, extracting and sorting feature point coordinates from a text centerline graph, determining a polygonal bounding box, and classifying feature points using text direction and boundary offset graphs, enabling text recognition in any direction with improved efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a two-stage solution of detection and recognition is used, then text detection and recognition can be performed separately with conventional methods, but the system complexity increases and real-time processing becomes difficult
Solution Approach 1:
The patent merges text detection and recognition into a single unified network structure. The detection head outputs text location information while the recognition head outputs text content, both processed simultaneously through shared feature extraction layers. This integration eliminates the need for separate detection and recognition stages, reducing system complexity while maintaining accuracy.
Solution Approach 2:
The patent segments the unified network into specialized components: a detection head for location information and a recognition head for text content. This segmentation allows each component to be optimized for its specific function while sharing the backbone feature extraction, resolving the contradiction between unified processing and specialized performance.
2Extent of automation
If a character-based method with large backbone network is used, then end-to-end detection and recognition can be achieved, but real-time processing becomes impossible due to computational requirements
Solution Approach 1:
The patent extracts only the essential feature extraction functionality into a compact backbone network, separating it from the heavier character-based processing. By using polygonal bounding boxes and direction offsets instead of full character-level analysis, the method maintains end-to-end capability while dramatically reducing computational requirements for real-time processing.
Solution Approach 2:
The patent changes the representation parameters from detailed character-level features to simplified polygonal bounding boxes and direction offsets. This parameter transformation reduces the dimensionality and complexity of the data processed by the backbone network, enabling real-time processing while preserving end-to-end detection and recognition functionality.
3Measurement precision
If conventional text detection methods are used, then text position can be determined, but text in any direction cannot be recognized accurately
Solution Approach 1:
The patent adds directional information as an additional dimension to the text representation. By introducing direction offset graphs and sorting feature points according to text direction, the system can accurately recognize and process text in any orientation while maintaining precise position determination through the polygonal bounding box framework.
Data Source
AI summary
The present disclosure discloses a method and apparatus for recognizing a text. The method comprises: acquiring images of a text area of an input image, the acquired images including a text centerline graph, a text direction offset graph, a text boundary offset graph, and a text character classification graph; extracting coordinates of feature points of a character center from the text centerline graph; sorting the extracted coordinates of the feature points based on the text direction offset graph to obtain a coordinate sequence of the feature points; determining a polygonal bounding box of the text area based on the coordinate sequence of the feature points of the character center and the text boundary offset graph; and determining a classification result of the feature points of the character center, based on the coordinate sequence of the feature points of the character center and the text character classification graph.