Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

UKDM: Underwater keypoint detection and matching using underwater image enhancement techniques

Pedro Diaz-Garcia, Felix Escalona, Miguel Cazorla

TL;DR

Underwater imaging hampers keypoint detection and matching due to light attenuation and turbidity. The paper evaluates multiple deep learning–based image enhancement networks (e.g., FSpiral-GAN, Funie-GAN) as preprocessing for SuperPoint/SuperGlue pipelines across two datasets, using metrics such as AUC, Precision, REP, and MScore, including uncertainty analyses. Results show that FSpiral-GAN and Funie-GAN significantly improve matching robustness, particularly with post-processing, while traditional detectors fall short in repeatability. The findings support a practical underwater SLAM/navigation workflow that leverages advanced image enhancement to achieve reliable feature detection and matching in challenging environments.

Abstract

The purpose of this paper is to explore the use of underwater image enhancement techniques to improve keypoint detection and matching. By applying advanced deep learning models, including generative adversarial networks and convolutional neural networks, we aim to find the best method which improves the accuracy of keypoint detection and the robustness of matching algorithms. We evaluate the performance of these techniques on various underwater datasets, demonstrating significant improvements over traditional methods.

UKDM: Underwater keypoint detection and matching using underwater image enhancement techniques

TL;DR

Underwater imaging hampers keypoint detection and matching due to light attenuation and turbidity. The paper evaluates multiple deep learning–based image enhancement networks (e.g., FSpiral-GAN, Funie-GAN) as preprocessing for SuperPoint/SuperGlue pipelines across two datasets, using metrics such as AUC, Precision, REP, and MScore, including uncertainty analyses. Results show that FSpiral-GAN and Funie-GAN significantly improve matching robustness, particularly with post-processing, while traditional detectors fall short in repeatability. The findings support a practical underwater SLAM/navigation workflow that leverages advanced image enhancement to achieve reliable feature detection and matching in challenging environments.

Abstract

The purpose of this paper is to explore the use of underwater image enhancement techniques to improve keypoint detection and matching. By applying advanced deep learning models, including generative adversarial networks and convolutional neural networks, we aim to find the best method which improves the accuracy of keypoint detection and the robustness of matching algorithms. We evaluate the performance of these techniques on various underwater datasets, demonstrating significant improvements over traditional methods.

Paper Structure

This paper contains 15 sections, 7 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Proposal
  4. Experimentation
  5. Datasets
  6. First dataset: Low Quality Videos
  7. Second dataset: Underwater Caves Sonar and Vision Dataset
  8. SuperGlue for Keypoint Matching
  9. SuperGlue in Underwater Image Matching
  10. Evaluation Metrics
  11. Analysis of Keypoint Matching Performance
  12. Uncertainty Test with FSpiral-GAN
  13. Uncertainty Analysis on other networks
  14. Keypoint Matching Results for the second dataset
  15. Conclusions

Figures (7)

  • Figure 1: Frame of one video of the first dataset.
  • Figure 2: Example of the underwater cave environment used for the Girona dataset.
  • Figure 3: SuperGlue feature matching between two enhanced frames. Color indicates confidence level, from low (blue) to high (red).
  • Figure 4: Results obtained on the uncertainity test for FSpiral-GAN without preprocesing
  • Figure 5: Results obtained on the uncertainity test for FSpiral-GAN with preprocesing
  • ...and 2 more figures