APSIPA Transactions on Signal and Information Processing > Vol 11 > Issue 1

ITS-Net: Iterative Two-Stream Network for Image Super-Resolution

Wei Li, School of Software, Shandong University, and IOT Research Center, China Electronics Standardization Institute, China, Yan Huang, School of Software, Shandong University, China, yan.h@sdu.edu.cn , Yilong Yin, School of Software, Shandong University, China, Jingliang Peng, Shandong Provincial Key Laboratory of Network Based Intelligent Computing, and School of Information Science and Engineering, University of Jinan, China
 
Suggested Citation
Wei Li, Yan Huang, Yilong Yin and Jingliang Peng (2022), "ITS-Net: Iterative Two-Stream Network for Image Super-Resolution", APSIPA Transactions on Signal and Information Processing: Vol. 11: No. 1, e29. http://dx.doi.org/10.1561/116.00000018

Publication Date: 21 Sep 2022
© 2022 W. Li, Y. Huang, Y. Yin and J. Peng
 
Subjects
 
Keywords
Image resolutionlow-frequencyhigh-frequencydeep learningiterative learning
 

Share

Open Access

This is published under the terms of CC BY-NC.

Downloaded: 127 times

In this article:
Introduction 
Related Work 
Our Approach 
Experiment 
Conclusion 
References 

Abstract

Remarkable progress on single image super-resolution (SISR) has been achieved with deep convolutional neural network (CNN) based approaches. These methods usually divide the images into high-frequency (HF) and low-frequency (LF) components and mainly recover the high-frequency component in a supervised manner. However, only simple interpolation manners, such as bilinear and bicubic, are utilized in the LF components recovery process, which limits the performance of these SISR approaches. We argue that the recovery of LF components also plays important roles in SISR, and to relieve the problem, we propose an iterative two-stream network (ITS-Net) which recovers the LF and HF components with convolution operations, respectively, thus better high-resolution images can be obtained. To achieve this, we utilize a sub-network with convolution and deconvolution operations to recover the LF components, and an iterative learning strategy is used to obtain well recovered LF and HF components. Extensive experiments on various benchmarking datasets demonstrate the effectiveness of our approach comparing with state-of-the-art CNN based approaches.

DOI:10.1561/116.00000018