Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing License Plate Super-Resolution: A Layout-Aware and Character-Driven Approach

Valfride Nascimento, Rayson Laroca, Rafael O. Ribeiro, William Robson Schwartz, David Menotti

TL;DR

A novel loss function, Layout and Character Oriented Focal Loss (LCOFL), which considers factors such as resolution, texture, and structural details, as well as the performance of the LPR task itself, outperforming two state-of-the-art methods in both quantitative and qualitative measures.

Abstract

Despite significant advancements in License Plate Recognition (LPR) through deep learning, most improvements rely on high-resolution images with clear characters. This scenario does not reflect real-world conditions where traffic surveillance often captures low-resolution and blurry images. Under these conditions, characters tend to blend with the background or neighboring characters, making accurate LPR challenging. To address this issue, we introduce a novel loss function, Layout and Character Oriented Focal Loss (LCOFL), which considers factors such as resolution, texture, and structural details, as well as the performance of the LPR task itself. We enhance character feature learning using deformable convolutions and shared weights in an attention module and employ a GAN-based training approach with an Optical Character Recognition (OCR) model as the discriminator to guide the super-resolution process. Our experimental results show significant improvements in character reconstruction quality, outperforming two state-of-the-art methods in both quantitative and qualitative measures. Our code is publicly available at https://github.com/valfride/lpsr-lacd

Enhancing License Plate Super-Resolution: A Layout-Aware and Character-Driven Approach

TL;DR

A novel loss function, Layout and Character Oriented Focal Loss (LCOFL), which considers factors such as resolution, texture, and structural details, as well as the performance of the LPR task itself, outperforming two state-of-the-art methods in both quantitative and qualitative measures.

Abstract

Despite significant advancements in License Plate Recognition (LPR) through deep learning, most improvements rely on high-resolution images with clear characters. This scenario does not reflect real-world conditions where traffic surveillance often captures low-resolution and blurry images. Under these conditions, characters tend to blend with the background or neighboring characters, making accurate LPR challenging. To address this issue, we introduce a novel loss function, Layout and Character Oriented Focal Loss (LCOFL), which considers factors such as resolution, texture, and structural details, as well as the performance of the LPR task itself. We enhance character feature learning using deformable convolutions and shared weights in an attention module and employ a GAN-based training approach with an Optical Character Recognition (OCR) model as the discriminator to guide the super-resolution process. Our experimental results show significant improvements in character reconstruction quality, outperforming two state-of-the-art methods in both quantitative and qualitative measures. Our code is publicly available at https://github.com/valfride/lpsr-lacd
Paper Structure (16 sections, 5 equations, 4 figures, 3 tables)

This paper contains 16 sections, 5 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. License Plate Recognition
  4. Super-Resolution for License Plate Recognition
  5. Proposed Approach
  6. Layout and Character Oriented Focal Loss
  7. Classification Loss
  8. LP Layout Penalty
  9. Dissimilarity Loss
  10. Architecture
  11. Experiments
  12. Setup
  13. Experimental Results
  14. Ablation Study
  15. Preliminary Experiments on Real-World Data
  16. ...and 1 more sections

Figures (4)

  • Figure 1: Some lp images from the RodoSol-ALPR dataset laroca2022cross. The first two rows show Brazilian lp, while the last two show Mercosur lp. This work focuses on lp with all characters arranged in a single row (10k images).
  • Figure 2: Examples of hr-lr image pairs used in our experiments.
  • Figure 3: Representative images produced by the proposed approach and baseline methods for the same inputs. GT = Ground Truth.
  • Figure 4: Super-resolved lp generated by our method and baselines from real-world images. The background image shows the original scene from which the lr image was extracted. From top to bottom: lp reconstructions by SR3 saharia2023image, PLNET nascimento2023super, our method, and a reference hr image from a different frame.