SNIDER: Single Noisy Image DEnoising and Rectification

for Improving License Plate Recognition

Younkwan Lee* Juhyun Lee* Hoyeon Ahn Moongu Jeon

Machine Learning and Vision Lab (MLV)

Gwangju Institute of Science and Technology (GIST)

Abstract

In this paper, we present an algorithm for real-world license plate recognition (LPR) from a low-quality image. Our method is built upon a framework that includes denoising and rectification, and each task is conducted by Convolutional Neural Networks. Existing denoising and rectification have been treated separately as a single network in previous research. In contrast to the previous work, we here propose an end-to-end trainable network for image recovery, Single Noisy Image DEnoising and Rectification (SNIDER), which focuses on solving both the problems jointly. It overcomes those obstacles by designing a novel network to address the denoising and rectification jointly. Moreover, we propose a way to leverage optimization with the auxiliary tasks for multi-task fitting and novel training losses. Extensive experiments on two challenging LPR datasets demonstrate the effectiveness of our proposed method in recovering the high-quality license plate image from the low-quality one and show that the the proposed method outperforms other state-of-the-art methods.

Overall Architecture

Figure 1. The proposed system consists of two components: single noisy image denoising and rectification (SNIDER) for recovering a low-quality license plate image and a license plate recognition (LPR) network for recognizing the final recovery image. The SNIDER is an end-to-end trainable network with auxiliary tasks for better image recovery. The LPR network uses a pre-trained DarkNet based on YOLO v3 to detect texts.

Experimental Results

1. Results on AOLP-RP

Table 1. Full LPR performance (percentage) comparison of our method with the existing methods on AOLP-RP

2. Results on VTLP

Table 2. Full LPR performance (percentage) comparison of our method with the existing methods on VTLP

3. Ablation Study

Figure 2. Ablation Study. (a) : shows noisy input; (b) : only contains denoising net; (c) : only contains rectification net; (d) : adds all main tasks; (e) : adds segment task from (d); (f) : adds counting task from (d); (g) : adds all of tasks, namely our proposed model.

Table 3. Ablation study on the effectiveness of different components. DSN, RSN, SD, and CD represent the GD, GR, Ds, and Dc, respectively.

4. Qualitative Results

Figure 3. Visual comparison of different license plate recovery methods: For the three sample images in the first column, columns 2-4 show the recovery images by using [35], [38] and SNIDER, respectively. The sample images are from VTLP which suffer from geometric distortions as well as low-quality. The proposed SNIDER performs better in LPR recovery.

Publication

SNIDER: Single Noisy Image Denoising and Rectification for Improving License Plate Recognition

Younkwan Lee*, Juhyun Lee*, Hoyeon Ahn and Moongu Jeon

In IEEE/CVF International Conference on Computer Vision (ICCV) Workshop, Seoul, Korea, 2019. (ICCVW 2019)

[PDF] [Bibtex] [Dataset] (* indicates equal contributions)

Downloads

■ VTLP Dataset (Coming Soon...)
■ Source Code [PyTorch] [TensorFlow] (Coming Soon...)

When you use the code or dataset, please cite above work. This software and dataset are free for NON-COMMERCIAL use.

Contact

Project administrator:
- Younkwan Lee (brightyoun@gist.ac.kr) - Juhyun Lee (leejuhyun@gist.ac.kr)

Google Sites

Report abuse