Character Image Grouping for More Accurate String Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for character recognition in images, such as CRAFT and STAR-Net, suffer from reduced accuracy due to improper clipping of character regions, leading to inefficiencies in recognizing character strings.
Innovation Solution
An information processing device and method that includes probability image acquisition, character region estimation, clustering, and character recognition, where images are classified into a plurality of groups based on estimated positions, and recognition target images are generated to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If character string region is clipped based on probability images, then processing efficiency is improved, but recognition accuracy is reduced
Solution Approach 1:
The patent divides the character string region into multiple candidate regions based on different clipping methods (horizontal clipping, vertical clipping, and diagonal clipping). Each clipping method generates separate candidate regions, which are then evaluated to select the most appropriate one for recognition. This segmentation approach allows the system to maintain processing efficiency while improving recognition accuracy by considering multiple potential regions.
Solution Approach 2:
The patent changes the clipping parameters by applying different clipping methods (horizontal, vertical, diagonal) with varying angles and directions. By adjusting these geometric parameters, the system generates diverse candidate regions that better capture the actual character string boundaries, thereby improving recognition accuracy without sacrificing processing efficiency.
2Measurement precision
If character string region includes more characters, then clustering accuracy is improved, but recognition complexity increases
Solution Approach 1:
The patent segments the character string region into multiple candidate regions using different clipping methods. Each candidate region contains a subset of characters that can be independently evaluated. This segmentation reduces recognition complexity by breaking down large character groups into manageable subsets while maintaining clustering accuracy through multiple evaluation perspectives.
Solution Approach 2:
The patent generates multiple candidate regions that may include slightly more or fewer characters than the ideal group size. By evaluating multiple partial regions with varying character counts, the system achieves accurate clustering without requiring perfect precision in each individual region, thereby reducing overall recognition complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The information processing device acquires a probability image representing a probability of an existence of a character in each of pixels included in a target image including a plurality of characters based on the target image, estimates positions of respective character images included in the target image based on the acquired probability image, classifies the plurality of character images into a plurality of groups based on the estimated positions, acquires a plurality of recognition target images which is generated so as to correspond to the plurality of groups, and includes the plurality of character images respectively belonging to the corresponding groups, and recognizes the plurality of characters from each of the recognition target images.