Fast simultaneous image super-resolution and motion deblurring with decoupled cooperative learning
- PDF / 3,197,343 Bytes
- 14 Pages / 595.276 x 790.866 pts Page_size
- 10 Downloads / 223 Views
SPECIAL ISSUE PAPER
Fast simultaneous image super‑resolution and motion deblurring with decoupled cooperative learning Heng Liu1 · Jiajun Qin1 · Zilin Fu1 · Xue Li1 · Jungong Han2 Received: 2 December 2019 / Accepted: 9 April 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract In recent years, deep convolutional neural networks (CNNs) have been widely applied to handle low-level vision problems. However, most existing CNN-based approaches can either handle single degeneration each time or treat them jointly through feature entangling, thus likely leading to poor performance when the actual degradation is inconsistent with hypothetical degradation condition. Furthermore, feature coupling will bring a large amount of computation, which may make the methods impractical to real-time mobile scenarios. In order to address these problems, we propose a deep decoupled cooperative learning model which can not only develop the corresponding recover network to deal with each degradation, but also flexibly handle multiple degradations at the same time. Thus, our approach can achieve disentangling and synthesizing single image super-resolution and motion deblurring, which has high practicability. We evaluate the proposed approach on various benchmark datasets, covering both natural images and synthetic images. The results demonstrate its superiority, compared to the state-of-the-art, where image SR and motion deblurring can be accomplished effectively concurrently. The source code of the work is available at https://github.com/hengliusky/Cooperative-Learning-Deblur-SR. Keywords Image super-resolution · Motion deblurring · Decoupled cooperative learning
1 Introduction As is well known, resolution reduction and motion blurring are two main manifestations of image degradation, in which the former is usually caused by down-sampling, while the latter arises when the image recorded within a single exposure changes due to rapid movement. In contrast to such two degradation processes, image super-resolution (SR) and motion deblurring are the corresponding reverse processes— reconstructing high resolution (HR) images from low resolution (LR) counterparts, and recovering sharp images from blurred ones. For single image SR (SISR), the goal of super-resolving a low-resolution (LR) image is to recover the missing highfrequency details in the original HR image. In theory, a typical resolution degradation model consists of a serial of * Heng Liu [email protected] 1
School of Computer Science and Technology, Anhui University of Technology, Ma’anshan 243032, China
Warwick Manufacturing Group, The University of Warwick, Conventry CV4 7AL, UK
2
degenerating operations, such as smoothing, down-sampling and adding additive noise, which can be denoted as:
y = (x ∗ k)↓s + n,
(1)
where x is the HR image and y is the LR image, ∗ represents the convolution (blur) operator, k denotes the blur (smoothing) kernel, ↓s indicates a down-sampling operator with factor s, and n refers to additive white Gaussian n
Data Loading...
