Learning Polarization Image Restoration with General Restoration Priors

2026-09-24

AI Summary

This study addresses the challenges of multiple coupled degradations in real-world polarization images and the limited generalization of existing methods by proposing a unified polarization restoration framework. The framework leverages normalized Stokes representations to decouple intensity from polarization information and constructs a dual-branch architecture based on a Mixture-of-Experts (MoE) model. Through cross-domain feature transformation and distillation mechanisms, it effectively transfers general priors into the polarization domain. Additionally, this work establishes the first benchmark dataset for polarization images under compound degradations. Experiments demonstrate that the proposed method significantly improves restoration quality for complexly degraded polarization images on both public and newly introduced benchmarks, thereby effectively supporting diverse downstream vision tasks.

Abstract

Polarization imaging captures distinctive surface and geometric cues that benefit a wide range of vision tasks. However, real-world polarization acquisition is often affected by multiple coupled degradations, making image restoration essential for practical polarization vision. Existing methods are largely tailored to specific degradations and remain constrained by the limited scale and quality of polarization data. To address these limitations, we develop an all-in-one polarization restoration framework for diverse and composite degradations. We first study the impact of different polarization representations on restoration performance and identify the normalized Stokes representation as an effective choice for separating intensity and polarization information. Accordingly, we devise a dual-branch architecture that separates intensity and polarization modeling. To overcome the limitations of polarization-specific training, the intensity branch leverages pretrained general restoration priors and a mixture-of-experts extension for composite degradations, while its restoration knowledge is adaptively distilled into the symmetric polarization branch via a cross-domain feature transform. In addition, we establish a composite-degradation polarization benchmark to support all-in-one restoration research. Extensive experiments on public datasets and our proposed benchmark demonstrate the effectiveness of the proposed method.

Problem

• polarization image restoration
• composite degradations
• all-in-one restoration
• general restoration priors

Innovation

• Polarization Image Restoration
• Normalized Stokes Representation
• Dual-branch Architecture
• General Restoration Priors
• Cross-domain Feature Transform