Rotation-Equivariant Self-Supervised Method in Image Denoising

2025-05-26

AI Summary

This work addresses self-supervised image denoising without paired clean images, introducing rotational equivariance as a prior into self-supervised frameworks for the first time. Methodologically: (1) we design a network architecture strictly satisfying rotational equivariance, employing high-precision equivariant convolutions and deriving a theoretical bound on equivariance error; (2) we propose a dual-stream adaptive masking fusion mechanism that dynamically integrates an equivariant branch with a conventional CNN branch. Our contribution lies in the first realization of network-level rotational equivariance modeling in self-supervised denoising, backed by theoretical guarantees. Extensive experiments demonstrate consistent and significant performance gains across three mainstream self-supervised paradigms—Noise2Self, Noise2Void, and a Noise2Noise variant—validating the substantive enhancement afforded by rotational equivariance priors to self-supervised learning.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Adversarial Attacks & 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
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Security and Privacy: Data transparency and provenance

Abstract

Self-supervised image denoising methods have garnered significant research attention in recent years, for this kind of method reduces the requirement of large training datasets. Compared to supervised methods, self-supervised methods rely more on the prior embedded in deep networks themselves. As a result, most of the self-supervised methods are designed with Convolution Neural Networks (CNNs) architectures, which well capture one of the most important image prior, translation equivariant prior. Inspired by the great success achieved by the introduction of translational equivariance, in this paper, we explore the way to further incorporate another important image prior. Specifically, we first apply high-accuracy rotation equivariant convolution to self-supervised image denoising. Through rigorous theoretical analysis, we have proved that simply replacing all the convolution layers with rotation equivariant convolution layers would modify the network into its rotation equivariant version. To the best of our knowledge, this is the first time that rotation equivariant image prior is introduced to self-supervised image denoising at the network architecture level with a comprehensive theoretical analysis of equivariance errors, which offers a new perspective to the field of self-supervised image denoising. Moreover, to further improve the performance, we design a new mask mechanism to fusion the output of rotation equivariant network and vanilla CNN-based network, and construct an adaptive rotation equivariant framework. Through extensive experiments on three typical methods, we have demonstrated the effectiveness of the proposed method.

Problem

• Incorporating rotation equivariance in self-supervised image denoising.
• Theoretical analysis of rotation equivariant convolution layers.
• Designing a fusion mechanism for rotation equivariant and CNN networks.

Innovation

• Uses rotation-equivariant convolution for denoising
• Introduces adaptive rotation equivariant framework
• Proposes new mask mechanism for network fusion