develop image restoration algorithms

Addressing blind image restoration under unknown real-world degradations without ground-truth images, this work proposes a holistic solution. First, we introduce a learnable degradation chain estimator to accurately model complex, realistic degradations. Second, we design a consistency-driven, plug-and-play diffusion prior framework enabling end-to-end lightweight optimization. Third, we pioneer reference-free proxy metrics—MSE and LPIPS computed on synthetically degraded samples—that overcome the longstanding challenge of unreliable performance evaluation in blind restoration. To our knowledge, this is the first work unifying degradation modeling, restoration algorithm design, and no-reference assessment within a single coherent pipeline. Extensive experiments demonstrate substantial improvements in ranking accuracy over SOTA methods across multiple blind restoration benchmarks, significantly enhancing algorithmic assessability, comparability, and practical applicability.

Existing image restoration methods suffer from limited generalization and low efficiency when handling unknown or composite degradations. To address this, this work proposes the RAR framework, which, for the first time, integrates image quality assessment (IQA) and image restoration (IR) into a unified end-to-end iterative process within a shared latent space, enabling joint optimization of degradation identification, image restoration, and quality verification. The method establishes a dynamic, adaptive all-in-one restoration mechanism that effectively minimizes inter-module information loss and latency. Extensive experiments demonstrate that RAR consistently achieves state-of-the-art performance across scenarios involving single, unknown, and composite degradations.

This paper addresses unsupervised video restoration and enhancement—aiming to improve visual quality and support downstream vision tasks—under the practical constraint of lacking paired ground-truth data. To tackle this challenge, it systematically surveys mainstream approaches grounded in domain adaptation, self-supervised signal design, and blind-spot networks, and introduces, for the first time, a multi-dimensional taxonomy encompassing model architectures, loss functions, and degradation modeling. It further proposes a synthetic-data-driven evaluation paradigm to uniformly benchmark existing methods across denoising, frame interpolation, and super-resolution. Key contributions include: (i) establishing a principled pathway for video quality enhancement without paired supervision; (ii) revealing the synergistic mechanism between noise modeling and structural priors; and (iii) providing a comprehensive theoretical framework and practical guidelines for future research. (149 words)

Real-world image degradations—such as scratches, fading, noise, and low resolution—cause severe detail loss and hinder object-level color control in restoration. Method: We propose Internal Image Detail Enhancement (IIDE), a novel approach that embeds latent-space degradation modeling and diffusion-based denoising guidance into a pre-trained Stable Diffusion model, enabling joint preservation of structural/texture fidelity and text-driven, object-level local coloring—without fine-tuning. Contribution/Results: By synergistically integrating generative priors with conditional control, IIDE achieves state-of-the-art performance both qualitatively and quantitatively (e.g., superior LPIPS and FID scores). Experiments demonstrate high-fidelity restoration coupled with professional-grade editability, establishing a new paradigm for restoring real-world degraded images.

In visual perception, image demosaicing under noise corruption has long suffered from quality degradation and texture distortion, exacerbated by the lack of a unified modeling framework. This paper proposes the first end-to-end joint demosaicing and denoising network. Its core innovation lies in a dual-branch discriminator architecture that synergistically integrates VGG-based perceptual loss and adversarial loss, guiding the generator to perform residual learning and perceptual optimization. The entire model is fully differentiable and requires no multi-stage training. Quantitatively, it achieves significant improvements over contemporary state-of-the-art methods across objective metrics (PSNR, LPIPS) and subjective user studies, while maintaining comparable computational cost. Qualitatively, reconstructed images exhibit more realistic textures and more natural noise suppression. This work establishes a scalable, jointly optimized paradigm for low-level vision tasks.

This work addresses the limitation of existing real-world image restoration methods, which rely on ground-truth supervision of inconsistent quality and tend to converge toward outputs with merely average perceptual quality. To overcome this, the authors propose IQPIR, a novel framework that explicitly incorporates no-reference image quality assessment (NR-IQA)-derived quality priors into the restoration process for the first time. By integrating these quality priors with a learned codebook prior through a quality-conditioned Transformer, a dual-branch discrete codebook architecture, and an optimized discrete representation strategy, IQPIR steers the model toward generating perceptually optimal results. The framework is plug-and-play—requiring no modification to the backbone network—and effectively disentangles generic from high-quality-specific features. It outperforms state-of-the-art methods on real image restoration benchmarks and functions as a general-purpose quality-guided module to enhance other restoration models.

This study systematically evaluates the capabilities and limitations of generative image restoration (GIR) methods in practical applications, revealing a shift in failure modes from under-generation to over-generation. To this end, we establish a multidimensional evaluation framework that comprehensively analyzes the performance of diffusion models, GANs, PSNR-oriented approaches, and general-purpose generative models across key dimensions including detail fidelity, sharpness, semantic correctness, and overall perceptual quality. Through large-scale subjective and objective experiments, we identify the central challenges as balancing fine-grained detail preservation with semantic controllability. Leveraging these insights, we develop a novel image quality assessment (IQA) model better aligned with human perception, offering a new benchmark and guiding direction for future GIR research.

This work addresses the challenges of task interference and high training costs in monolithic models for multi-degradation image restoration by proposing a modular, task-decoupled unified framework. The approach employs a lightweight CNN-based router to diagnose the degradation type of an input image and dynamically routes it to a dedicated U-Net expert model for on-demand restoration. Its key innovation lies in an explicit diagnosis-and-routing mechanism that enables model-agnostic, flexible scalability: incorporating a new degradation type requires training only a single expert module and fine-tuning the router, without retraining the entire system. Experimental results demonstrate that the proposed framework significantly reduces training overhead on standard hardware, avoids feature interference among tasks, and achieves superior performance compared to complex monolithic architectures.

This work addresses the limitation of existing unified image restoration methods, which typically apply a single strategy globally and struggle to handle spatially varying degradation types and severities. To overcome this, we propose MGN-AIR, a novel framework that introduces, for the first time, a pixel-level multimodal guidance mechanism. By fusing textual semantics with visual cues, MGN-AIR enables precise perception and adaptive restoration of local degradations at the pixel level. The framework comprises three key components: pixel-wise visual prompt estimation, multimodal prompt fusion, and a prompt-guided restoration network, collectively enhancing fine-grained control over the restoration process. Extensive experiments demonstrate that MGN-AIR consistently outperforms state-of-the-art methods across multiple tasks, including denoising, deraining, deblurring, dehazing, desnowing, and low-light enhancement.

研究通过对比RAW和sRGB域的图像恢复方法，发现ISP转换意识下的RGB恢复模型表现最佳，强调了恢复模型与成像管道匹配的重要性。