Abstract:Image denoisers have been shown to be powerful priors for solving inverse problems in imaging. In this work, we introduce a generalization of these methods that allows any image restoration network to be used as an implicit prior. The proposed method uses priors specified by deep neural networks pre-trained as general restoration operators. The method provides a principled approach for adapting state-of-the-art restoration models for other inverse problems. Our theoretical result analyzes its convergence to a stationary point of a global functional associated with the restoration operator. Numerical results show that the method using a super-resolution prior achieves state-of-the-art performance both quantitatively and qualitatively. Overall, this work offers a step forward for solving inverse problems by enabling the use of powerful pre-trained restoration models as priors.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is: **How to use the pre - trained image inpainting network as an implicit prior to solve the inverse problems in imaging**. Specifically, the paper proposes a new method - Deep Restoration Priors (DRP), which allows any image inpainting network to be used as an implicit prior, thus providing a general framework for solving inverse problems. ### Detailed Explanation: 1. **Background and Motivation**: - Many problems in computational imaging, biomedical imaging, and computer vision can be formulated as inverse problems, that is, recovering high - quality images from low - quality observed data. - These inverse problems are usually ill - posed, so prior models are required to accurately infer the unknown images. - Current methods are mainly based on deep learning (DL), where deep models are trained to map observed data to images. - Image denoisers have been widely used to specify image priors to solve inverse problems. However, existing work mainly focuses on utilizing the statistical properties of denoisers, while less attention is paid to other types of image inpainting operators (such as super - resolution models). 2. **Core Contributions of the Paper**: - **Proposing the DRP Method**: This method allows the use of any pre - trained image inpainting network as an implicit prior, expanding the application scope of existing methods. - **Theoretical Analysis**: Provides a theoretical convergence analysis of DRP under the minimum mean - square - error (MMSE) estimator, showing that it can be interpreted as minimizing a composite objective function containing the logarithm of the degraded image density. - **Experimental Verification**: By using the popular SwinIR super - resolution model as a prior, it demonstrates the potential advantages of DRP in two different inverse problems, deblurring and super - resolution, proving that it can achieve state - of - the - art performance. 3. **Solutions to Specific Problems**: - The DRP method provides a principled framework by combining pre - trained inpainting networks and measurement models, enabling powerful pre - trained inpainting models to be used to solve other inverse problems. - The paper verifies the effectiveness of DRP through numerical experiments, especially when using super - resolution priors, DRP performs better than traditional image denoisers. ### Summary: The main purpose of this paper is to fill the gaps in the existing literature. By proposing the DRP method, pre - trained image inpainting networks can be used as an effective implicit prior to solve various inverse problems. This not only expands the application scope of existing methods but also provides new ideas for future theoretical research and practical applications.

A Restoration Network as an Implicit Prior

Stochastic Deep Restoration Priors for Imaging Inverse Problems

FiRe: Fixed-points of Restoration Priors for Solving Inverse Problems

Image Restoration using Autoencoding Priors

BIGPrior: Towards Decoupling Learned Prior Hallucination and Data Fidelity in Image Restoration

Equivariant Denoisers for Image Restoration

A Set-Theoretic Study of the Relationships of Image Models and Priors for Restoration Problems

Variational Image Restoration Network

Variational Image Restoration Network.

Plug-and-Play Image Restoration with Deep Denoiser Prior

Image Restoration Based on End-to-End Unrolled Network

Integrating Local and Non-local Denoiser Priors for Image Restoration.

Denoising Prior Driven Deep Neural Network for Image Restoration

Restoration by Generation with Constrained Priors

Deep unfolding of a proximal interior point method for image restoration

Effective Image Restorations Using a Novel Spatial Adaptive Prior

A Deep Optimization Approach for Image Deconvolution.

Multi-Channel and Multi-Model-Based Autoencoding Prior for Grayscale Image Restoration

Deep Denoiser Prior Based Deep Analytic Network for Lensless Image Restoration.

Image Restoration Via Joint Low-Rank and External Nonlocal Self-Similarity Prior

Learning Deep CNN Denoiser Prior for Image Restoration