End-to-end license plate correction and recognition method

Abstract

The invention discloses an end-to-end license plate correction and recognition method, and belongs to the technical field of image processing. The method comprises the steps of obtaining a license plate image, and taking the license plate image as the input of a license plate correction and recognition fusion model which comprises a backbone network, a license plate correction head and a license plate character recognition head, wherein the license plate correction head and the license plate character recognition head share the backbone network; the backbone network performs multi-scale low-level and high-level feature extraction and fusion on the license plate image to obtain a license plate feature map F; the license plate correction head performs license plate correction based on the license plate feature map F to obtain a corrected license plate; and the license plate character recognition head recognizes license plate characters based on the license plate feature map F. According to the invention, the license plate correction step and the license plate recognition step are fused into an end-to-end deep learning model, so that the recognition precision of a difficult license plate is improved while the license plate correction step is simplified.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an end-to-end license plate correction and recognition method. Background technique [0002] License plate recognition is one of the core technologies of smart cities, and it has a wide range of applications in scenarios such as entrance and exit billing, illegal capture, and vehicle tracking. [0003] In recent years, the development momentum of license plate recognition algorithms has been good, and the deep learning methods have been fully popularized, which has greatly improved the recognition accuracy. However, due to the lack of rotation invariance of the convolutional network, at the same time, the default license plate direction is horizontal when the commonly used license plate recognition network is designed, such as the CTC series of methods, making the existing license plate recognition methods difficult to deal with severe distortion, large angle tilt, etc. ...

AI Technical Summary

Problems solved by technology

[0004] At present, there are two mainstream algorithms for license plate correction: one is based on traditional image processing methods such as color segmentation, edge detection, and line detection to determine the correction matrix of the license plate image; this type of method is better for simple samples, but for difficult Sample robustness is poor

The other is a method based on deep learning, which directly predicts the correction matrix of the license plate image through the convolutional network; this method has better performance and strong generalization ability, but requires a large number of license plate corners to annotate images, and will increase additional computing power requirement

Images