Deep Learning-Driven Ultra-High-Definition Image Restoration: A Survey

TL;DR

This survey systematically maps the landscape of deep learning-driven UHD image restoration, detailing degradation models, UHD-specific challenges, and a broad set of UHD benchmark datasets. It classifies methods into three architectural paradigms—downsampling-enhancement-upsampling, encoder-decoder with stepwise up-downsampling, and resampling-enhancement—and catalogs progress across SR, dehazing, LLIE, deblurring, deraining, and desnowing. It also analyzes technical developments in network design, sampling strategies, and loss functions, and provides performance comparisons with computational complexity considerations. By proposing a classification framework and highlighting real-world data gaps, the work lays out concrete directions for lightweight models, priors-based guidance, and specialized UHD evaluation metrics to advance practical UHD restoration. The accompanying repository offers a resource for reproducibility and further exploration of UHD restoration research.

Abstract

Ultra-high-definition (UHD) image restoration aims to specifically solve the problem of quality degradation in ultra-high-resolution images. Recent advancements in this field are predominantly driven by deep learning-based innovations, including enhancements in dataset construction, network architecture, sampling strategies, prior knowledge integration, and loss functions. In this paper, we systematically review recent progress in UHD image restoration, covering various aspects ranging from dataset construction to algorithm design. This serves as a valuable resource for understanding state-of-the-art developments in the field. We begin by summarizing degradation models for various image restoration subproblems, such as super-resolution, low-light enhancement, deblurring, dehazing, deraining, and desnowing, and emphasizing the unique challenges of their application to UHD image restoration. We then highlight existing UHD benchmark datasets and organize the literature according to degradation types and dataset construction methods. Following this, we showcase major milestones in deep learning-driven UHD image restoration, reviewing the progression of restoration tasks, technological developments, and evaluations of existing methods. We further propose a classification framework based on network architectures and sampling strategies, helping to clearly organize existing methods. Finally, we share insights into the current research landscape and propose directions for further advancements. A related repository is available at https://github.com/wlydlut/UHD-Image-Restoration-Survey.

Figures (12)

• Figure 1: Examples of UHD images under different degradation conditions. These images often suffer from blur, haze, low light, rain, and snow.
• Figure 2: Milestones in UHD image restoration methods are categorized into three primary structures: Downsampling-Enhancement-Upsampling, Encoder-Decoder with Stepwise Up-downsampling, and Resampling-Enhancement.
• Figure 3: Taxonomy of deep learning-based UHD image restoration across datasets, approaches, and techniques.
• Figure 4: Summary of the downsampling-enhancement-upsampling structure for UHD image restoration. (a) The single-branch downsampling-enhancement-upsampling architecture focuses on the design of enhancement networks in the low-resolution space, utilizing some popular architectures such as CNN, UNet, MLP, and Transformer. (b) The dual-branch downsampling-enhancement-upsampling architecture explores the correlation between high- and low-resolution features or incorporates additional prior information to guide the reconstruction process.
• Figure 5: Overview of dual branch frameworks under the downsampling-enhancement-upsampling structure. DMixer dmixer upsamples low-resolution features and merges them with high-resolution features for reconstruction. UDR-Mixer UDRMixer feeds high-resolution features into the low-resolution branch to facilitate reconstruction. UHDformer uhdformer transforms features from high to low resolution and enhances high-resolution reconstruction through concatenation. UHDDIP uhddip extracts gradient and normal priors in the low-resolution space to interact with low-resolution features, guiding high-resolution reconstruction.