Text Image Character Segmentation Using Convex Hulls
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional character segmentation methods in text image processing are limited by issues such as slanted characters being segmented as a whole, limited applicability to specific text environments, and inability to handle stroke fracture or cohesion phenomena, resulting in low versatility and accuracy.
Innovation Solution
The method involves preprocessing text images to obtain binary images with connected regions, using convex hull algorithms to segment characters, and merging character blocks based on their heights to improve segmentation accuracy and versatility, preventing up-down structured characters from being incorrectly segmented.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If projection segmentation method is used, then character segmentation can be performed on binary images, but slanted characters are segmented as a whole reducing accuracy
Solution Approach 1:
The patent applies segmentation by dividing the character region into multiple scanning lines and further segmenting connected regions on each scanning line into character blocks based on projection values, enabling accurate separation of slanted characters that conventional methods fail to separate
Solution Approach 2:
The patent introduces a new dimension by using scanning lines (horizontal dimension) in addition to the traditional vertical projection method, allowing characterization of slanted characters through both horizontal and vertical projection values, thereby resolving the limitation of handling slanted characters
2Adaptability or versatility
If clustering method based on connected regions is used, then character blocks can be merged based on distribution features, but stroke fracture or cohesion phenomena prevent proper segmentation
Solution Approach 1:
The patent changes parameters by introducing projection value thresholds and scanning line intervals as adjustable parameters, allowing the system to adapt to different text distributions and stroke characteristics, thereby handling both fracture and cohesion phenomena effectively
Solution Approach 2:
The patent applies dynamics by making the segmentation process adaptive through adjustable thresholds and parameters that can be optimized based on specific text characteristics, enabling the system to dynamically adjust to different scenarios including stroke fracture and cohesion
3Measurement precision
If template matching method is used, then specific fonts or characters can be recognized, but applicability is limited to specific text environments
Solution Approach 1:
The patent applies universality by creating a segmentation method that works across different fonts, sizes, and text environments through projection-based characterization rather than template-specific matching, making the system universally applicable while maintaining high accuracy
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
A text image processing method and apparatus. The method comprises: preprocessing a text image to obtain a binary image and multiple connected domains comprised in the binary image (210); obtaining, by means of a convex hull algorithm, convex hulls corresponding respectively to the multiple connected domains and character regions externally connected to the convex hulls (230); performing horizontal-direction character segmentation on the obtained character regions to obtain multiple character blocks distributed in the binary image (250); and merging the character blocks according the heights of the character blocks in the binary image to obtain blocks comprised in the text image (270). The text image processing method and apparatus can improve the versatility and the accuracy of character segmentation.