Kernel Adversarial Learning for Real-world Image Super-resolution

• DOI: 10.48550/arxiv.2104.09008
• Publication Date: 2021-01-01
• Authors: Hu Wang; Congbo Ma; Jianpeng Zhang; Gustavo Carneiro
• Keywords: Kernel (algebra)

Current deep image super-resolution (SR) approaches attempt to restore high-resolution images from down-sampled or by assuming degradation simple Gaussian kernels and additive noises. However, such processing techniques represent crude approximations of the real-world procedure lowering resolution. In this paper, we propose a more realistic process lower resolution introducing new Kernel Adversarial Learning Super-resolution (KASR) framework deal with SR problem. proposed framework, noises are adaptively modeled rather than explicitly specified. Moreover, also an iterative supervision high-frequency selective objective further boost model reconstruction accuracy. Extensive experiments validate effectiveness on datasets.