Exploring the Transformer-Based and Diffusion-Based Models for Single Image Deblurring

作     者：Park, Seunghwan Shin, Chaehun Lew, Jaihyun Yoon, Sungroh 

作者机构：Data Science and AI Laboratory ECE and Interdisciplinary Program in AI Seoul National University Seoul Korea Republic of Large Display Development Center Samsung Display Corporation Yongin Korea Republic of 

出 版 物：《SSRN》 

年 卷 期：2024年

核心收录：

主　　题：Convolutional neural networks 

摘      要：Image deblurring is a fundamental task in image restoration (IR) aimed at removing blurring artifacts caused by factors such as defocusing, motions, and others. Since a blurry image could be originated from various sharp images, deblurring is regarded as an ill-posed problem with multiple valid solutions. The evolution of deblurring techniques spans from rule-based algorithms to deep learning-based models. Early research focused on estimating blur kernels using maximum a posteriori (MAP) estimation, but advancements in deep learning have shifted the focus toward directly predicting sharp images. The performance of deblurring models was significantly improved by leveraging deep learning techniques such as convolutional neural networks (CNNs), generative adversarial networks (GANs), recurrent neural networks (RNNs), and others. Especially, models utilizing Transformers and Vision Transformers (ViTs) demonstrate state-of-the-art (SOTA) performance. The advent of diffusion probabilistic models (DPMs) has further enhanced this progress by leveraging their generative capabilities for image restoration tasks. We provide an in-depth investigation of representative deblurring models, from traditional approaches to state-of-the-art deep learning models, categorizing them according to their foundational technologies. In addition, a detailed evaluation of representative models using standardized metrics is also presented, highlighting advances in both methodology and practical performance. © 2024, The Authors. All rights reserved.