StreamUR: Physics-informed Near Real-Time Underwater Image Restoration

The exploration of underwater environments poses significant challenges due to the optical properties of water, leading to color distortion, reduced contrast and blurring in images. This work aims to enhance the clarity and fidelity of underwater images and videos in near real-time. The SeaThru phys...

Full description

Saved in:
Bibliographic Details
Published in:International archives of the photogrammetry, remote sensing and spatial information sciences. remote sensing and spatial information sciences., 2024-11, Vol.XLVIII-3-2024, p.1-8
Main Authors: Antoniou, Christos, Spanos, Sotiris, Vellas, Simon, Ntouskos, Valsamis, Karantzalos, Konstantinos
Format: Article
Language:English
Subjects:
Autonomous underwater vehicles
Color
Datasets
Image contrast
Image enhancement
Image processing
Image quality
Image restoration
Optical properties
Physics
R&D
Real time
Remotely operated vehicles
Research & development
Underwater
Underwater construction
Unmanned vehicles
Video
Visual fields
Visual observation
Visual tasks
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The exploration of underwater environments poses significant challenges due to the optical properties of water, leading to color distortion, reduced contrast and blurring in images. This work aims to enhance the clarity and fidelity of underwater images and videos in near real-time. The SeaThru physics-based color correction method was suitably adapted for obtaining target images across a diverse collection of underwater datasets considered. Based on these target images, the MIMO-UNet model is used to address the processing speed limitations of the physics-based correction methods, enabling near real-time image and video processing without explicit depth information. The proposed method has been integrated into autonomous underwater observation systems and remotely operated vehicle (ROV) cameras, offering enhanced visibility. Additionally, we build a MIMO-UNet network for generating realistic synthetic underwater images, valuable for training and simulation. This research advances underwater imaging enhancement and restoration, significantly improving visual data quality and vision-dependent tasks in submerged environments. The public release of the dataset aims to facilitate further research and development in this field.
ISSN:2194-9034
1682-1750
2194-9034
DOI:10.5194/isprs-archives-XLVIII-3-2024-1-2024