Title: Image super-resolution algorithm based on V-transform combined with neural network
Authors: Nan Nan; Shijie Yue; Ruipeng Gang; Chenghua Li; Ruixia Song
Addresses: Department of Science, North China University of Technology, Beijing, Shijingshan, China ' Department of Science, North China University of Technology, Beijing, Shijingshan, China ' NRTA, Academy of Broadcasting Science, Beijing, China ' Institute of Automation, Academy of Sciences, Beijing, China ' Department of Science, North China University of Technology, Beijing, Shijingshan, China
Abstract: Single image super resolution aims to increase the visual quality and size of low-resolution images. Although the existing deep learning-based methods have achieved promising results, there is still room for improvement in both subjective and objective effects due to the inability of deep convolution to balance low-frequency content and high-frequency details in the reconstruction of complex scenes. To solve this problem, we propose a V-transform based image super resolution model combined with convolution (VTSR). The VTSR is mainly composed of three parts: the V-transform block, feature fusion model and the upsampling module. Test experiments on four standard data sets show that our proposed V-transform based image super resolution model combined with convolution can achieve better results than most methods at all scales. And the three innovations we propose have positive effects on super resolution tasks to varying degrees.
Keywords: single image super resolution; CNN; the V-transform; frequency domain.
DOI: 10.1504/IJCAT.2023.132399
International Journal of Computer Applications in Technology, 2023 Vol.71 No.4, pp.296 - 310
Received: 28 May 2022
Received in revised form: 24 Jul 2022
Accepted: 26 Jul 2022
Published online: 19 Jul 2023 *