[Paper Review] Image Denoising: The Deep Learning Revolution and Beyond -- A Survey Paper --

Michael Elad, Bahjat Kawar+1 more|arXiv (Cornell University)|Jan 9, 2023

Cell Image Analysis Techniques9 citations

TL;DR

This survey recounts the evolution of image denoising from classical Bayesian and signal-processing approaches to modern deep-learning-based methods, and explores how denoisers now serve broader roles in inverse problems and image synthesis.

ABSTRACT

Image denoising (removal of additive white Gaussian noise from an image) is one of the oldest and most studied problems in image processing. An extensive work over several decades has led to thousands of papers on this subject, and to many well-performing algorithms for this task. Indeed, 10 years ago, these achievements have led some researchers to suspect that "Denoising is Dead", in the sense that all that can be achieved in this domain has already been obtained. However, this turned out to be far from the truth, with the penetration of deep learning (DL) into image processing. The era of DL brought a revolution to image denoising, both by taking the lead in today's ability for noise removal in images, and by broadening the scope of denoising problems being treated. Our paper starts by describing this evolution, highlighting in particular the tension and synergy that exist between classical approaches and modern DL-based alternatives in design of image denoisers. The recent transitions in the field of image denoising go far beyond the ability to design better denoisers. In the 2nd part of this paper we focus on recently discovered abilities and prospects of image denoisers. We expose the possibility of using denoisers to serve other problems, such as regularizing general inverse problems and serving as the prime engine in diffusion-based image synthesis. We also unveil the idea that denoising and other inverse problems might not have a unique solution as common algorithms would have us believe. Instead, we describe constructive ways to produce randomized and diverse high quality results for inverse problems, all fueled by the progress that DL brought to image denoising. This survey paper aims to provide a broad view of the history of image denoising and closely related topics. Our aim is to give a better context to recent discoveries, and to the influence of DL in our domain.

Motivation & Objective

• Define the image denoising problem and its ill-posed nature.
• Review the historical development of priors and denoisers from the classical era to the deep-learning era.
• Explain how deep learning reshapes denoising practice and its relation to AI advances.
• Discuss how denoisers are used as building blocks for regularizing inverse problems and for image generation/synthesis.

Proposed method

• Present the problem formulation for AWGN denoising and MMSE/MAP estimators.
• Trace the evolution of image priors from simple regularizers to sparsity and low-rank models.
• Describe classical denoising techniques (e.g., BM3D, WNNM) and their principles.
• Outline the DL-based denoising paradigm: data-driven training, noise modeling, and loss design.
• Discuss the Plug-and-Play and Regularization by Denoising concepts and diffusion-model-based synthesis.

Experimental results

Research questions

• RQ1 How has the priors’ design evolved from classical to learning-based approaches in image denoising?
• RQ2 What impact has deep learning had on the performance and scope of image denoisers compared to traditional methods?
• RQ3 How can denoisers be leveraged as priors or regularizers for other inverse problems?
• RQ4 What prospects do denoisers offer for randomized, diverse, high-perceptual-quality inverse problem solutions and image synthesis?

Key findings

• Deep-learning denoisers now lead in noise suppression, outperforming many classical methods.
• Denoisers can serve as powerful priors/regularizers for inverse problems, enabling new solution strategies.
• Denoisers enable randomized, diverse, high perceptual quality results for inverse problems, revealing uncertainty.
• Denoising progress intertwines with advancements in diffusion models and perceptual quality optimization.
• The survey highlights synergy and tension between classical priors and AI-based designs in denoising.