Not All Degradations Are Equal: A Targeted Feature Denoising Framework for Generalizable Image Super-Resolution

2025-09-18

AI Summary

Generic image super-resolution (SR) models often overfit degradation artifacts—particularly noise—under unknown degradations, rather than learning intrinsic content features, thereby compromising generalization. This work is the first to identify noise as the primary source of overfitting in SR under arbitrary degradations. To address this, we propose the Targeted Feature Denoising (TFD) framework, comprising a lightweight noise detection module and a feature-level denoising module. TFD requires no architectural modification to the backbone network and can be seamlessly integrated as a plug-and-play component into existing SR models, while remaining fully compatible with conventional regularization techniques. Extensive experiments on five benchmark datasets demonstrate that TFD consistently outperforms state-of-the-art regularization-based generalization methods, achieving superior reconstruction quality under both synthetic and real-world multi-degradation scenarios. These results validate TFD’s effectiveness, universality, and robustness across diverse degradation conditions.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Generalizable Image Super-Resolution aims to enhance model generalization capabilities under unknown degradations. To achieve this goal, the models are expected to focus only on image content-related features instead of overfitting degradations. Recently, numerous approaches such as Dropout and Feature Alignment have been proposed to suppress models' natural tendency to overfit degradations and yield promising results. Nevertheless, these works have assumed that models overfit to all degradation types (e.g., blur, noise, JPEG), while through careful investigations in this paper, we discover that models predominantly overfit to noise, largely attributable to its distinct degradation pattern compared to other degradation types. In this paper, we propose a targeted feature denoising framework, comprising noise detection and denoising modules. Our approach presents a general solution that can be seamlessly integrated with existing super-resolution models without requiring architectural modifications. Our framework demonstrates superior performance compared to previous regularization-based methods across five traditional benchmarks and datasets, encompassing both synthetic and real-world scenarios.

Problem

• Addressing model overfitting to noise in super-resolution
• Proposing targeted feature denoising for unknown degradations
• Enhancing generalization across synthetic and real-world images

Innovation

• Targeted feature denoising framework
• Noise detection and denoising modules
• Seamless integration with existing models