A Dataset and Model for Realistic License Plate Deblurring

TL;DR

This work tackles the challenge of license plate deblurring under realistic motion blur. It introduces LPBlur, a large-scale paired dataset collected with a dual-camera system to obtain aligned sharp and blurred license plate images, and LPDGAN, a GAN-based deblurring model featuring multi-scale latent fusion, a text reconstruction module, and a partition discriminator to improve per-letter details. The approach achieves superior deblurring performance and improves license plate recognition accuracy compared with state-of-the-art methods, including robust performance in low-light conditions. The dataset and model together offer a practical resource for training and evaluating realistic license plate deblurring systems, with potential impact on intelligent traffic management and automated recognition tasks.

Abstract

Vehicle license plate recognition is a crucial task in intelligent traffic management systems. However, the challenge of achieving accurate recognition persists due to motion blur from fast-moving vehicles. Despite the widespread use of image synthesis approaches in existing deblurring and recognition algorithms, their effectiveness in real-world scenarios remains unproven. To address this, we introduce the first large-scale license plate deblurring dataset named License Plate Blur (LPBlur), captured by a dual-camera system and processed through a post-processing pipeline to avoid misalignment issues. Then, we propose a License Plate Deblurring Generative Adversarial Network (LPDGAN) to tackle the license plate deblurring: 1) a Feature Fusion Module to integrate multi-scale latent codes; 2) a Text Reconstruction Module to restore structure through textual modality; 3) a Partition Discriminator Module to enhance the model's perception of details in each letter. Extensive experiments validate the reliability of the LPBlur dataset for both model training and testing, showcasing that our proposed model outperforms other state-of-the-art motion deblurring methods in realistic license plate deblurring scenarios. The dataset and code are available at https://github.com/haoyGONG/LPDGAN.

Figures (6)

• Figure 1: The visual deblurring results of several state-of-the-art models and our model for real-world motion blurred license plate images.
• Figure 2: (a) A schematic diagram of the paired image acquisition system collecting data in a pedestrian bridge. (b) The pipeline of paired images post-processing.
• Figure 3: Overview of the proposed Licence Plate Deblurring Generative Adversarial Network.
• Figure 4: The architecture of Partition Discriminator Module.
• Figure 5: Visual comparison of different deblurring methods on LPBlur dataset. The test scene is divided into normal light and low light. Since the results in low light scenes are difficult to distinguish visually, we uniformly increase their brightness. The original brightness of the low light scene refers to the first line, which is Blur Image.