License Plate Character Recognition Using Residual CNNs Without Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional license plate recognition systems face challenges in ensuring accuracy due to difficulties in character segmentation, leading to lower recognition performance.
Innovation Solution
A method utilizing a convolutional neural network with a residual network structure and a bidirectional recurrent neural network model for direct license plate character recognition, avoiding gradient vanishing and feature loss, and eliminating the need for character segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If character segmentation is performed on license plate images, then the recognition process can be broken down into manageable parts, but the segmentation difficulties lead to lower recognition accuracy
Solution Approach 1:
The patent merges the character segmentation step with the recognition step by using a end-to-end deep learning model that directly processes the entire license plate image and outputs recognized characters, eliminating the need for separate segmentation and thereby improving accuracy while maintaining manageable complexity through modular network architecture
Solution Approach 2:
The patent applies segmentation at the network architecture level rather than image processing level, dividing the recognition task into parallel processing streams (e.g., separate branches for different character types or positions) that are later combined, allowing the system to handle complexity internally while presenting a unified interface
2Productivity
If conventional recognition methods are used, then the system structure is simpler, but the recognition speed and accuracy are lower
Solution Approach 1:
The patent replaces traditional mechanical image processing steps (segmentation, feature extraction, classification) with a unified deep learning model that performs all tasks through learned representations, achieving higher speed and accuracy at the cost of increased network complexity that is managed through architectural design choices
3Productivity
If the entire license plate is processed without segmentation, then recognition speed improves, but character identification becomes more difficult
Solution Approach 1:
The patent transitions from processing individual characters in isolation to processing the entire license plate image as a unified structure, adding the dimension of spatial context and relationships between characters, which makes character identification easier through contextual information while maintaining high speed through efficient network design
Data Source
AI summary
A license plate character recognition method and apparatus, an electronic device, and a storage medium. The method comprises: after obtaining a vehicle image captured by an image capturing device, positioning a vehicle plate area in the vehicle image to obtain a license plate image, and performing feature extraction on the license plate image by means of a convolutional neural network comprising a residual network structure. Gradient vanishing is effectively avoided and feature loss in a convolutional process of the convolutional neural network is reduced, so that a bidirectional recurrent neural network model can accurately recognize license plate characters in the license plate image on the basis of feature information of the license plate image. Therefore, character segmentation for a license plate is not needed, license plate characters on the license plate can be obtained by directly recognizing the whole license plate, segmentation and separate recognition for the license plate characters are avoided, and the recognition speed and the recognition accuracy are improved.