Single Image Super-resolution Reconstruction Based on Enhanced Residual Network. Issue 1 (February 2021)

• Record Type: Journal Article
• Authors: Liu, Chunyu; Qian, Wenhua; Xu, Dan; Jiang, Mengjie; Li, Xiaojin
• Is Part Of: Journal of physics. Volume 1815:Issue 1(2021)
• Publication Date: 2021-02
• Subjects: Image super-resolution -- Fusion -- Residual learning -- Deconvolution; Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1815/1/012015
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: High-resolution images present richer detailed information and have stronger information expression capabilities. The increase of the network depth does not guarantee that the reconstructed image has a higher quality, and may cause problems such as overfitting. So this article proposes an enhanced residual network, which can fully extract input low-resolution image features and reconstruct high-resolution images. This paper introduces a deconvolution operation based on the residual module to expand the size of input features, and the connection between different modules promotes feature fusion, obtains more high-frequency details from the input low-resolution image. The objective experimental results show that the proposed method has improved the indicators PSNR and SSIM. In terms of visual effects, it can reconstruct clearer and more detailed images.