Character Coordinate Extraction via Feature Fusion Heat Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting character coordinates in images, particularly in mixed texts, face challenges such as determining touching characters and achieving robustness due to variations in character widths and the absence of closed areas in common characters.
Innovation Solution
A method that inputs a target text image into a feature extraction backbone network, fuses character and text line segmentation features, and uses these features to calculate individual character coordinates through character and text line segmentation heat maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If character width-based segmentation is used to determine touching characters, then segmentation can be performed for uniform-width texts, but it fails for mixed Chinese-English texts where character widths vary
Solution Approach 1:
The patent changes the detection parameter from character width (which varies between Chinese and English) to the presence of closed areas (which is a consistent structural feature). This parameter transformation allows the method to handle both single-byte and multi-byte characters uniformly, resolving the adaptability issue while maintaining reliable touching character detection.
2Reliability
If center position of closed area is used to segment touching characters, then touching characters can be segmented, but most common characters without closed areas cannot be processed
Solution Approach 1:
The patent makes the closed area detection serve multiple purposes: it is used both for detecting touching characters and for segmenting individual characters. By utilizing the same structural feature (closed areas) for both detection and segmentation tasks, the method achieves universal applicability across different character types while maintaining reliable segmentation of touching characters.
3Measurement precision
If attention mechanism-based recognition model is used for character segmentation, then recognition results can be obtained, but attention drift occurs affecting both recognition and segmentation accuracy
Solution Approach 1:
The patent separates the recognition function and segmentation function into independent processing streams. Instead of using attention mechanism for both tasks simultaneously (which causes attention drift), the method segments characters based on closed area detection independently from the recognition process, eliminating the negative interaction while maintaining both recognition accuracy and segmentation robustness.
Data Source
AI summary
Embodiments of the present application disclose a character coordinate extraction method and apparatus, a device, a medium and a program product. The method comprises: inputting a target text image into a feature extraction backbone network, and obtaining character segmentation features and text line segmentation features by means of feature fusion by different layers in the backbone network; respectively inputting the character segmentation features and the text segmentation features into a character segmentation module and a text line segmentation module, and obtaining a character segmentation heat map and a text segmentation heat map of the target text image, wherein the character segmentation module and the text line segmentation module form a segmentation network model; and calculating coordinates of a single character in the target text image according to the character segmentation heat map and the text line segmentation heat map. According to the embodiments of the present application, repeated extraction of features is reduced; high robustness is achieved for character segmentation; convergence of the network is accelerated, and the segmentation efficiency of the network is improved; the accuracy of single-character coordinate extraction is improved.