Equivariant Learning for Unsupervised Image Dehazing

2026-01-20

AI Summary

This work proposes EID, an unsupervised image dehazing framework tailored for scientific imaging, where the absence of ground-truth clear images and reliable priors poses significant challenges. EID uniquely integrates equivariant learning with physical haze modeling by leveraging image signal symmetries, haze consistency constraints, and system equivariance, enabling high-quality restoration without requiring clean reference images or handcrafted priors. To further capture the unknown haze formation process, adversarial learning is incorporated into the framework. Extensive experiments demonstrate that EID substantially outperforms existing methods across diverse benchmarks, including cellular microscopy, medical endoscopy, and natural images, thereby confirming its strong generalization capability and superior dehazing performance in complex scientific imaging scenarios.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Data transparency and provenance
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Image Dehazing (ID) aims to produce a clear image from an observation contaminated by haze. Current ID methods typically rely on carefully crafted priors or extensive haze-free ground truth, both of which are expensive or impractical to acquire, particularly in the context of scientific imaging. We propose a new unsupervised learning framework called Equivariant Image Dehazing (EID) that exploits the symmetry of image signals to restore clarity to hazy observations. By enforcing haze consistency and systematic equivariance, EID can recover clear patterns directly from raw, hazy images. Additionally, we propose an adversarial learning strategy to model unknown haze physics and facilitate EID learning. Experiments on two scientific image dehazing benchmarks (including cell microscopy and medical endoscopy) and on natural image dehazing have demonstrated that EID significantly outperforms state-of-the-art approaches. By unifying equivariant learning with modelling haze physics, we hope that EID will enable more versatile and effective haze removal in scientific imaging. Code and datasets will be published.

Problem

• Image Dehazing
• Unsupervised Learning
• Scientific Imaging
• Haze Removal

Innovation

• equivariant learning
• unsupervised image dehazing
• haze consistency
• adversarial learning
• scientific imaging