Blind Super Resolution with Reference Images and Implicit Degradation Representation

2025-07-18

AI Summary

Blind super-resolution (BSR) suffers from strong coupling between degradation kernels and scaling factors, making it challenging to model scale-aware degradation processes. Method: This paper proposes a decoupled degradation modeling framework leveraging external high-resolution (HR) reference images. For the first time, content-agnostic HR references are introduced; an implicit degradation representation mechanism adaptively generates scale-aligned low-resolution–high-resolution (LR–HR) pairs, explicitly decoupling degradation kernels from scaling factors. The method integrates multi-scale feature extraction with a joint degradation estimation network, enabling zero-shot transfer and compatibility with diverse BSR backbone architectures. Contribution/Results: Our approach achieves significant improvements over state-of-the-art methods on both synthetic and real-world blind degradation benchmarks, with substantial PSNR and SSIM gains. It demonstrates strong generalization across unseen degradation types and scaling factors, highlighting its practical applicability and robustness.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Multimodal Learning
• Search and Optimization: Learning to Search

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: Graph neural networks and deep learning approaches for Web-related graphs
• Security and Privacy: Large-scale security measurements

Abstract

Previous studies in blind super-resolution (BSR) have primarily concentrated on estimating degradation kernels directly from low-resolution (LR) inputs to enhance super-resolution. However, these degradation kernels, which model the transition from a high-resolution (HR) image to its LR version, should account for not only the degradation process but also the downscaling factor. Applying the same degradation kernel across varying super-resolution scales may be impractical. Our research acknowledges degradation kernels and scaling factors as pivotal elements for the BSR task and introduces a novel strategy that utilizes HR images as references to establish scale-aware degradation kernels. By employing content-irrelevant HR reference images alongside the target LR image, our model adaptively discerns the degradation process. It is then applied to generate additional LR-HR pairs through down-sampling the HR reference images, which are keys to improving the SR performance. Our reference-based training procedure is applicable to proficiently trained blind SR models and zero-shot blind SR methods, consistently outperforming previous methods in both scenarios. This dual consideration of blur kernels and scaling factors, coupled with the use of a reference image, contributes to the effectiveness of our approach in blind super-resolution tasks.

Problem

• Estimating scale-aware degradation kernels for blind super-resolution
• Utilizing HR reference images to improve degradation modeling
• Enhancing SR performance via adaptive LR-HR pair generation

Innovation

• Uses HR reference images for scale-aware kernels
• Adaptively discerns degradation with content-irrelevant references
• Generates LR-HR pairs via downsampling HR references