Confusable Character Recognition via Gradient Direction Changes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image recognition systems struggle to accurately identify confusable characters on license plates, especially when the plates are inclined or smeared, leading to incorrect recognition due to similar appearances of characters like 8 and B, 0 and D, etc.
Innovation Solution
A method and apparatus that recognize detected character images by calculating the number of changes in gradient for confusable characters, determining final character information based on these calculations, and setting feature areas to improve accuracy, thereby reducing incorrect recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image recognition is used to identify license plate characters, then the recognition process is simple and fast, but the accuracy is low when characters are confusable (e.g., 8 and B, 0 and D) especially under inclined or smeared conditions
Solution Approach 1:
The patent divides the character image into multiple gradient calculation directions (horizontal, vertical, and diagonal gradients). By segmenting the gradient analysis into different directional components, the system can capture more discriminative features of confusable characters, thereby improving recognition accuracy without requiring a complete redesign of the recognition architecture.
Solution Approach 2:
The patent transforms the character recognition problem from direct pixel comparison to gradient domain analysis. By calculating gradient magnitude and direction at each pixel, and then analyzing the number of gradient direction changes, the system changes the parameter space to one where confusable characters exhibit distinct patterns, improving discrimination capability.
2Reliability
If the license plate is inclined or smeared, then the character appearance becomes more similar and harder to distinguish, but the physical conditions cannot be controlled
Solution Approach 1:
The patent moves from analyzing raw pixel intensities to analyzing gradient directions in multiple dimensions. By computing gradients in horizontal, vertical, and diagonal directions, and then counting the number of gradient direction changes, the system adds dimensional information that remains stable under inclination and smear transformations, thereby improving reliability.
Solution Approach 2:
The patent changes from using absolute pixel values to using gradient direction change counts as recognition features. This parameter transformation makes the recognition system invariant to affine transformations such as inclination and smear, as the gradient direction change pattern remains consistent even when the character shape is distorted.
3Measurement precision
If more complex recognition methods are applied to distinguish confusable characters, then accuracy improves, but processing time and computational load increase
Solution Approach 1:
The patent extracts only the essential discriminative feature - the number of gradient direction changes - from the complex character image. By focusing on this single key feature rather than analyzing all pixel variations, the system achieves good discrimination of confusable characters with relatively simple computation, maintaining processing efficiency.
Solution Approach 2:
The patent transforms the recognition task into counting gradient direction changes, which is a computationally efficient operation compared to full image matching or deep learning approaches. This parameter transformation enables accurate confusable character discrimination with low computational overhead, preserving productivity.
Data Source
AI summary
An identification method and apparatus of confusable character are provided. The method involves: the detected character image is identified to gain the initial character information which is corresponding to the character image; the step change times of the corresponding external outline of the character image are counted if the initial character information is the confusable character; the final character information corresponding to the character image is confirmed according to the step change times; The final character information of the character image can be known conveniently according to the step change times, therefore the corresponding correct character information of the character image can be identified more precisely. The possibility of wrong identification of the character image because of the appearing confusable character can be reduced, and the identification precision rate of the confusable character can be improved.