Looking Into the Water by Unsupervised Learning of the Surface Shape

2026-03-08

AI Summary

This work addresses the challenge of image distortion in aerial observation of underwater scenes caused by refraction at the dynamic air-water interface. The authors propose an unsupervised image restoration method that, for the first time, leverages SIREN-based implicit neural representations for spatiotemporal modeling of the water surface. Their architecture employs a dual neural field network to jointly predict the water surface height and its spatial derivatives, along with the corresponding undistorted underwater image colors. Notably, the framework operates without ground-truth labels, enabling simultaneous reconstruction of sharp underwater image sequences and estimation of dynamic water surface geometry. Experimental results demonstrate that the proposed method outperforms current state-of-the-art unsupervised restoration approaches on both synthetic and real-world datasets, achieving high-fidelity recovery of both image content and water surface structure.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Machine Learning: Unsupervised & Self-Supervised Learning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

We address the problem of looking into the water from the air, where we seek to remove image distortions caused by refractions at the water surface. Our approach is based on modeling the different water surface structures at various points in time, assuming the underlying image is constant. To this end, we propose a model that consists of two neural-field networks. The first network predicts the height of the water surface at each spatial position and time, and the second network predicts the image color at each position. Using both networks, we reconstruct the observed sequence of images and can therefore use unsupervised training. We show that using implicit neural representations with periodic activation functions (SIREN) leads to effective modeling of the surface height spatio-temporal signal and its derivative, as required for image reconstruction. Using both simulated and real data we show that our method outperforms the latest unsupervised image restoration approach. In addition, it provides an estimate of the water surface.

Problem

• water surface refraction
• image distortion
• underwater imaging
• unsupervised learning
• neural fields

Innovation

• unsupervised learning
• neural fields
• water surface reconstruction
• SIREN
• refraction correction