Towards maximizing feature efficiency: All-in-one image restoration via radial basis attention

Wide-field nailfold capillary image deblurring method based on an improved MIMO-UNet.

HDFNet:Hybrid-domain fusion network for medical image restoration

An image restoration framework based on a multiscale transform and neural networks for one-dimensional rotating sparse aperture imaging systems

LIRNet: Boosting the performance for unified low-light image restoration

Compressive sensing image restoration with deep prior guided group sparse representation

Blind super-resolution (SR) aims to reconstruct high-resolution images from low-quality inputs under unknown degradation conditions. While numerous blind SR methods have been proposed in recent years, they still face critical limitations. Most approaches perform well under specific degradation patterns but struggle with complex scenarios involving multiple degradation factors and varying noise levels. This often leads to loss of structural integrity and fine details, resulting in suboptimal restoration quality. Furthermore, existing methods typically rely on convolutional neural networks (CNNs) with limited receptive fields, which hinders effective cross-domain information integration. Their inability to capture long-range dependencies compromises the reconstruction of global structures. In edge detection, conventional techniques frequently produce inaccurate or false edges, further degrading the quality of cross-domain integration and image restoration. To address these challenges, we propose Structural Integrity and Texture Fidelity Transformer (SITFFormer), a novel transformer-based framework for blind single-image SR. Our approach incorporates the Canny edge detection algorithm to accurately preserve true edges and suppress noise-induced artifacts, enhancing edge localization in complex and noisy environments. We also introduce the Cross-Domain Structure-Texture-Aware Network (CDSTNet), designed to integrate intra-domain and cross-domain features for comprehensive structure preservation and texture recovery. CDSTNet comprises two key modules: Cross-Domain Integration (CDI) that fuses intra- and cross-domain features to retain structural and textural details. Cross-Domain Learnable Attention (CDLA) that explores global dependencies, adaptively refines feature similarity, and filters out redundant non-local information. Both modules are equipped with a Cross-Attention Mechanism (CAM) to facilitate effective interaction and complementarity between domains, enhancing reconstruction fidelity. Extensive experiments on synthetic, noisy, and real-world datasets demonstrate that SITFFormer surpasses state-of-the-art methods in quantitative performance and visual quality, particularly in preserving structural integrity and recovering fine textures.

Modern image restoration and super-resolution methods utilize deep learning due to its superior performance compared to traditional algorithms. However, deep learning typically requires large labeled training datasets, which are rarely available in astrophotography. Deep Image Prior (DIP) bypasses this constraint by performing unsupervised optimization on a single image without training data; however, DIP often suffers from overfitting, artifact generation, and instability. This work proposes DIPLI - a framework designed specifically for resolved, high-contrast astronomical targets that shifts from single-frame to multi-frame processing using the Back Projection technique, combined with dense optical flow estimation via the TVNet model, and replaces deterministic predictions with Monte Carlo estimation obtained through Stochastic Gradient Langevin Dynamics (SGLD). A comprehensive evaluation compares the method against the original DIP, the transformer-based model RVRT, and the diffusion-based model DiffIR2VR-Zero on synthetic data with ground truth, while comparing qualitatively against Lucky Imaging on real astronomical data. On synthetic datasets, DIPLI achieves the best perceptual fidelity scores (LPIPS in 12/12 and DISTS in 10/12 scenarios), while the diffusion-based DiffIR2VR-Zero achieves the best pixel-level distortion scores (PSNR in 9/12 and SSIM in 8/12 scenarios), consistent with the well-known perceptual-distortion trade-off in image restoration(Blau and Michaeli In Proceedings of the IEEE Conference on Computer Visionand Pattern Recognition 6228-6237 2018). Compared to classical Lucky Imaging, the model requires far fewer input frames (7-13 versus thousands) and avoids the need for early stopping that limits standard DIP. Qualitative evaluation on real-world data of resolved solar-system objects, where ground truth is unavailable and domain shifts typically hinder generalization, suggests that the method appears to preserve fine detail while suppressing noise and artifacts.

Diffractive neural networks are a promising framework for all-optical processing of visual data, with the potential to drastically reduce the computational burden and energy consumption that is currently associated with running neural networks on digital hardware. Recently, diffractive networks have been applied to various computational imaging tasks. However, while these problems are commonly ill-posed, existing diffractive network designs output only a single reconstruction for each input image, and thus do not inform the user of the inherent uncertainty in the reconstruction. In this work, we explore a passive all-optical diffractive network architecture, together with a dedicated training loss, which allows the network to simultaneously output multiple plausible reconstructions for each input image. We numerically illustrate the efficacy of our method on the tasks of spatial super-resolution and imaging beyond opaque occluders. As we show, the set of diverse outputs generated by the network provides a highly informative visualization of the uncertainty in the reconstruction. Our approach is a first step towards unlocking the full potential of passive all-optical processing in scientific and/or safety-critical image reconstruction applications.

Abstract In real-world scenarios, the performance of semantic segmentation often deteriorates when processing low-quality (LQ) images, which may lack clear semantic structures and high-frequency details. Although image restoration techniques offer a promising direction for enhancing degraded visual content, conventional real-world image restoration (Real-IR) models primarily focus on pixel-level fidelity and often fail to recover task-relevant semantic cues, limiting their effectiveness when directly applied to downstream vision tasks. Conversely, existing segmentation models trained on high-quality data lack robustness under real-world degradations. In this paper, we propose Restoration Adaptation for Semantic Segmentation (RASS), which effectively integrates semantic image restoration into the segmentation process, enabling high-quality semantic segmentation on the LQ images directly. Specifically, we first propose a Semantic-Constrained Restoration (SCR) model, which injects segmentation priors into the restoration model by aligning its cross-attention maps with segmentation masks, encouraging semantically faithful image reconstruction. Then, RASS transfers semantic restoration knowledge into segmentation through LoRA-based module merging and task-specific fine-tuning, thereby enhancing the model’s robustness to LQ images. To validate the effectiveness of our framework, we construct a real-world LQ image segmentation dataset with high-quality annotations, and conduct extensive experiments on both synthetic and real-world LQ benchmarks. The results show that SCR and RASS significantly outperform state-of-the-art methods in segmentation and restoration tasks. Code, models, and datasets will be available at https://github.com/Ka1Guan/RASS.git .

Abstract Restoration of heritage images is very important since the degradation of cultural objects may be attributed to various factors such as background noise, blurring effects, missing parts, and so forth. In this study, PCA, CNN, and transformers were compared. Experiments showed that CNN enhances local features, whereas transformers enhance global features, while PCA serves as a baseline for noise removal. Hybrid models based on CNN and transformers have proved to produce better results in historical restoration projects. Digital restoration of historical images plays a very vital role in heritage preservation efforts. PCA, CNN, and transformers have improved accuracy restoration. But challenges such as computational complexity, real-time processing, and perception quality persist. This study offers suggestions for future research. Keywords Heritage Image Restoration, PCA, CNN, Transformer, Deep Learning, Hybrid, Cultural Heritage

In Transformer-based image restoration models, the self-attention mechanism often introduces attention noise from irrelevant contextual feature, hindering the recovery of underlying clear content. Although many methods have been proposed to suppress attention noise, we note that most existing approaches are often developed for general vision tasks and fail to generalize across remote sensing image dehazing, where large-scale spatial structures pose additional challenges for attention modeling. How to effectively model scale-aware attention to suppress redundant activations becomes crucial for remote sensing image dehazing. In this paper, we propose a scale-adaptive differential Transformer (SDTformer), an architecture designed to suppress attention noise through a differential attention mechanism, thereby improving reconstruction fidelity. Specifically, the model incorporates a scale-adaptive differential self-attention module, which models contextual dependencies across different spatial scales and reduces redundant contextual interference by computing differential attention maps. Additionally, a dynamic differential feed-forward network is proposed to adaptively select informative spatial features, strengthening feature aggregation. To further enhance feature representation, a gated fusion module is introduced to aggregate multi-scale features generated by different encoder blocks, which facilitates the learning process of each decoder block and improves the final reconstruction performance. Extensive experimental results on the commonly used benchmarks show that our method achieves favorable performance against state-of-the-art approaches.

X-ray imaging technology, as the core non-invasive inspection method, plays an irreplaceable role in industrial non-destructive testing and medical diagnosis. However, during signal acquisition, the imaging system faces multiple interferences, such as the quantum effect and electronic noise. This leads to a significant decrease in the image's signal-to-noise ratio, seriously affecting the accuracy of hazardous material identification and lesion detection. Existing X-ray image denoising methods have two major limitations. First, in physical model-driven denoising methods, the existing noise models deviate significantly from realistic ones, resulting in poor denoising results. Second, in mainstream deep learning-based methods, Convolutional Neural Networks (CNNs) have limitations in capturing long-range dependencies, while the Transformer model with a global receptive field has high computational complexity. To address these challenges, a physically grounded noise model is designed for synthesizing realistic X-ray images, trained on the public mainstream X-ray image security inspection datasets and augmented with hybrid real-synthetic data. Based on this, a novel denoising model, XDenoiser, is proposed in this paper. It incorporates a linear attention complexity Receptance Weighted Key-Value (RWKV) into a Transformer-based image restoration structure and combines it with CNNs to support both global and local receptive fields. Experiments on the expanded mainstream X-ray image security inspection datasets demonstrate the reasonableness and effectiveness of the XDenoiser algorithm.

Image restoration is a vital research area in computer vision, focusing on reconstructing high-quality clear images from degraded observations. Common types of degradation include noise and blur, which may stem from imaging device limitations, environmental interference, and other factors. This paper centers on the design and optimization of multi-stage image restoration networks, conducting in-depth exploration of feature extraction, feature fusion, attention mechanisms, and their practical applications. A multi-stage hybrid attention mechanism-based image restoration network is proposed. Initially, each stage progressively extracts and restores image features. Then, an adaptive feature fusion block enables effective cross-stage information transfer. Finally, by calculating losses at each stage and assigning different weights, the network achieves stable convergence during training. The hybrid attention mechanism enhances the model's focus on critical features and improves its understanding of the overall image structure. Outstanding performance has been achieved in both image deblurring and denoising tasks. On the GoPro dataset, the restored results achieved a PSNR of 33.26 and an SSIM of 0.963. On the SIDD dataset, the restored results reached a PSNR of 40.23 and an SSIM of 0.963. Furthermore, ablation experiments demonstrated the effectiveness of the multi-stage model, hybrid attention mechanism, and adaptive feature fusion block.

Person re-identification (ReID) under diverse weather conditions remains a critical yet insufficiently explored problem. Most existing ReID approaches are developed and benchmarked on clear-weather datasets, resulting in significant performance degradation when deployed in rainy, snowy, or hazy environments. Conventional image restoration methods, typically optimized for low-level image quality metrics, are often misaligned with the objectives of high-level identity discrimination and thus fail to improve the person ReID performance. To address these limitations, we propose DW-ReID, a unified framework that integrates weather-degraded image restoration with person re-identification tasks. The proposed DW-ReID is built upon a large-scale Contrastive Language-Image Pre-training (CLIP) model and achieved by a two-stage training paradigm. In the first stage, a set of learnable text prompts is optimized to construct identity-specific ambiguous descriptions for each person's identity. In the second stage, the optimized text descriptions, together with a frozen text encoder, provide language supervision to jointly train a weather encoder, an image restorer, and a ReID encoder in an end-to-end manner. The experimental results on two our contributed synthetic datasets consistently demonstrate the effectiveness and superior performance of the proposed DW-ReID method.

Universal restoration of medical images.

Historical images as the dominant method for documenting the world and its inhabitants can help us to better understand the real history. Due to the limited camera technology, historical images captured in the early to mid-20th century tend to be very blurry, unclear, noisy, and obscure. The goal of this paper is to super-resolve images for historical image restoration. Compared to the degradations in modern digital imagery, those in historical images have unique features that are typically much more complex and less well understood. The discrepancy between historical images and modern high-definition digital images leads to a significant performance drop for existing super-resolution (SR) models trained on modern digital imagery. To tackle this problem, we propose a new method, namely DA-CycleGAN. Specifically, the DA-CycleGAN is built on top of CycleGAN to achieve unsupervised learning. We introduce a degradation-adaptive (DA) module with strong, flexible adaptation to learn various unknown degradations from samples. Moreover, we collect a large dataset containing 10,000 low-resolution images from real historical films. The dataset features various natural degradations. Our experimental results demonstrate the superior performance of DA-CycleGAN and the effectiveness of our image dataset for achieving accurate super-resolution enhancement of historical images.

EFAIR: Efficient Hybrid Networks with Contrastive Compact Prompts for All-in-One Image Restoration

This study presents a machine learning approach for converting X-rayimages acquired with a thick-phosphor (low-resolution) detector intothose expected from a thin-phosphor (high-resolution) detector. Unlikeconventional super-resolution tasks, the focus is on restoring thedetector response rather than merely increasing the pixel density.Image degradation from blurring was modeled as a diffusion-like process,characterized by a gradual loss of spatial resolution, with the forwardprocess implemented as a progressive degradation of the modulation-transferfunction (MTF) from a high-resolution image. The inverse processwas approximated using a learned deep neural network. Performancewas evaluated in the spatial domain using the peak signal-to-noiseratio and structural similarity index and in the frequency domainusing the MTF, noise-power spectrum, and contrast-transfer function.The proposed method outperformed the conventional Wiener filtering,U-Net, and conditional diffusion models in both spatial- and frequency-domainanalyses. The stochastic nature of blurring from secondary quantumscattering was discussed, and modern diffusion-based generative modelswere proposed as promising frameworks for deblurring and image restoration.

Generalization of deep learning image restoration method for compressed sensing in electron tomography with a limited number of projections