Gradual deep residual network for super-resolution

• DOI: 10.1007/s11042-020-10152-9
• Publication Date: 2020-11-13T14:03:38Z
• Authors: Zhaoyang Song; Xiaoqiang Zhao; Hongmei Jiang
• Keywords: Upsampling; Benchmark (surveying)

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 low-resolution 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.