A New Dataset and Framework for Real-World Blurred Images Super-Resolution

Rui Qin,Ming Sun,Chao Zhou,Bin Wang

2024-07-20

Abstract:Recent Blind Image Super-Resolution (BSR) methods have shown proficiency in general images. However, we find that the efficacy of recent methods obviously diminishes when employed on image data with blur, while image data with intentional blur constitute a substantial proportion of general data. To further investigate and address this issue, we developed a new super-resolution dataset specifically tailored for blur images, named the Real-world Blur-kept Super-Resolution (ReBlurSR) dataset, which consists of nearly 3000 defocus and motion blur image samples with diverse blur sizes and varying blur intensities. Furthermore, we propose a new BSR framework for blur images called Perceptual-Blur-adaptive Super-Resolution (PBaSR), which comprises two main modules: the Cross Disentanglement Module (CDM) and the Cross Fusion Module (CFM). The CDM utilizes a dual-branch parallelism to isolate conflicting blur and general data during optimization. The CFM fuses the well-optimized prior from these distinct domains cost-effectively and efficiently based on model interpolation. By integrating these two modules, PBaSR achieves commendable performance on both general and blur data without any additional inference and deployment cost and is generalizable across multiple model architectures. Rich experiments show that PBaSR achieves state-of-the-art performance across various metrics without incurring extra inference costs. Within the widely adopted LPIPS metrics, PBaSR achieves an improvement range of approximately 0.02-0.10 with diverse anchor methods and blur types, across both the ReBlurSR and multiple common general BSR benchmarks. Code here: <a class="link-external link-https" href="https://github.com/Imalne/PBaSR" rel="external noopener nofollow">this https URL</a>.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in blind image super - resolution (BSR) techniques, current methods perform poorly when dealing with images with blurring effects. Although existing BSR methods perform well on general images, their effectiveness drops significantly when applied to image data containing blurs. Blur is very common in real - world images and is often intentionally used in photography to enhance visual perception, such as by blurring the background to highlight foreground objects. However, most existing BSR methods ignore the processing of such intentional blurs and often over - enhance the blurred areas, which impairs the actual perceived quality of the image.

To further study and solve this problem, the authors developed a new super - resolution dataset specifically for blurred images - the Real - world Blur - kept Super - Resolution (ReBlurSR) dataset, as well as a new BSR framework - the Perceptual - Blur - adaptive Super - Resolution (PBaSR) framework. The ReBlurSR dataset contains nearly 3,000 out - of - focus and motion - blurred image samples with different blur sizes and intensities. The PBaSR framework consists of two main modules: the Cross Disentanglement Module (CDM) and the Cross Fusion Module (CFM). These modules are designed to improve the performance of the framework when dealing with blurred and non - blurred data without incurring additional inference and deployment costs.