Gradual deep residual network for super -resolution
- PDF / 1,686,890 Bytes
- 14 Pages / 439.37 x 666.142 pts Page_size
- 32 Downloads / 210 Views
Gradual deep residual network for super-resolution Zhaoyang Song 1 & Xiaoqiang Zhao 1,2,3
& Hongmei Jiang
1,2,3
Received: 2 July 2020 / Revised: 22 September 2020 / Accepted: 23 October 2020 # Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract
Deep neural networks with single upsampling have achieved the improvement of performance for single image super-resolution. However, these networks lose a lot of details of lowresolution image in the reconstruction process. In this paper, we propose a gradual deep residual network for super-resolution (GDSR), which consists of multiple reconstruction network with 2 scale factor (2X reconstruction network). In 2X reconstruction network, a residual block connected by residual (RBR) is proposed to form a deep residual network, which is used to extract the depth features of low-resolution images; then the extracted features are upsampled into the features of high-resolution image by sub-pixel convolutional layer. GDSR gradually reconstructs high-quality high-resoluiton images from low-resolution images by multiple 2X reconstruction networks. Extensive experiments on benchmark datasets demonstrate that the proposed GDSR outperforms the state-of-the-art methods in terms of quantitative evaluation, visual evaluation, and execution time evaluation. Keywords Single image super-resolution . Gradual deep residual network . Residual block connected by residual . Sub-pixel convolutional layer
1 Introduction Image super-resolution reconstruction refers to the recovery of high-resolution (HR) image from low-resolution (LR) image, which is widely used in satellite imaging [5, 38], urban surveilance [41], and medical [30] fields. Due to the increasing requirements in many application fields, image super-resolution reconstruction has become a research hotspot. Since
* Xiaoqiang Zhao [email protected]
1
College of Electrical Engineering and Information Engineering, Lanzhou University of Technology, Lanzhou 730050, China
2
Key Laboratory of Gansu Advanced Control for Industrial Processes, Lanzhou 730050, China
3
National Experimental Teaching Center of Electrical and Control Engineering, Lanzhou University of Technology, Lanzhou 730050, China
Multimedia Tools and Applications
some detail information of low-resolution image is lost, image super-resolution reconstruction is essentially an ill-posed problem. Although many important progresses have been made in past decade [6, 11, 27, 36], how to reconstruct better high-resolution image is still a fundamental and challenging task. In the past decade, super-resolution reconstruction methods were divided into three categories: interpolation-based methods, reconstruction-based methods, and learning-based methods [16]. Interpolation-based methods are the most basic image super-resolution reconstruction methods. Although interpolation-based methods can achieve good results, they are prone to problems such as blurring and ringing effects. Reconstruction-based methods are also classic super-resolution reconstruction met
Data Loading...
