Deep ResNet Based Remote Sensing Image Super-Resolution Reconstruction in Discrete Wavelet Domain
- PDF / 1,222,187 Bytes
- 10 Pages / 612 x 792 pts (letter) Page_size
- 0 Downloads / 232 Views
Deep ResNet Based Remote Sensing Image Super-Resolution Reconstruction in Discrete Wavelet Domain Q. Qina,*, J. Doua,**, and Z. Tua,*** a
Department of Automation and Mechanical and Electrical engineering, School of Intelligent Manufacturing and Control Engineering, Shanghai Second Polytechnic University, Shanghai, 201209 China * e-mail: [email protected] ** e-mail: [email protected] *** e-mail: [email protected]
Abstract—We present a single-image super-resolution (SR) method for Remote Sensing Image based on deep learning within Discrete Wavelet Domain in this paper. Our method is inspired Residual Learning. Firstly, an input image is decomposed by single level 2D Discrete wavelet transform to get four sub-bands. The four sub-bands coefficients are feeding into the Deep Learning Residual Network to predict correspondingly residual images; Adding four sub-band images and residual images as the new sub-bands of 2D wavelet transform; Finally, uses the inverse 2D Discrete wavelet transform to get the final output Super Resolution HR image. Our proposed method performs better than existing methods in accuracy and visual improvements in our results are easily noticeable. Keywords: Super Resolution, Discrete Wavelet Transform, Deep Learning, Convolutional Neural Network, Residual Learning DOI: 10.1134/S1054661820030232
INTRODUCTION Single image super-resolution (SISR) [1, 2], which aims at recovering a high-resolution (HR) image from a single low resolution (LR) image. Since SISR restores the high-frequency information, it is widely used in applications such as medical imaging [3], satellite imaging [4], security and surveillance [5], where high-frequency details are greatly desired. SISR is an inherently ill-posed problem since a multiplicity of solutions exist for any given low-resolution pixel. Such a problem is typically mitigated by constraining the solution space by strong prior information. In order to learn the prior, recent state-ofthe-art methods mostly adopt the example-based [6] strategy. These methods either explored the similarity of self-examples [7, 8], while others [9, 10] mapped the LR to HR patches with use of external samples. For the remote sensing and satellite imaging applications, wavelet transform is often adopted to solve the SISR problem to enhance image resolution for subsequently processing. Discrete wavelet transform (DWT) [11] is one of the wavelet transforms which is used in image processing. DWT decomposes an image into different sub-band images, namely low-low (LL), low-high (LH), high-low (HL), and high-high (HH). Stationary wavelet transform (SWT) [12] is a more
Received October 10, 2018; revised February 14, 2020; accepted February 21, 2020
efficient implementation then the two-dimensional discrete wavelet transform (DWT), which can obtain shift invariance. In recent years, due to the powerful learning ability, Deep Learning (DL) models, especially Convolutional Neural Networks (CNN), are widely used to address the ill-posed inverse problem of Super Resolution (SR), and have demo
Data Loading...
