2021/03/25 by Kai Zhang, Zhang, Kai, Jingyun Liang +5 · 65 citations
Computer Science · Engineering · #Advanced Image Processing Techniques #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Infrared Target Detection Methodologies #Optical Systems and Laser Technology #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2103.14006
openalex publication_date 2021/03/25 · openalex created_date 2022/10/02 · openalex updated_date 2026/07/28
It is widely acknowledged that single image super-resolution (SISR) methods\nwould not perform well if the assumed degradation model deviates from those in\nreal images. Although several degradation models take additional factors into\nconsideration, such as blur, they are still not effective enough to cover the\ndiverse degradations of real images. To address this issue, this paper proposes\nto design a more complex but practical degradation model that consists of\nrandomly shuffled blur, downsampling and noise degradations. Specifically, the\nblur is approximated by two convolutions with isotropic and anisotropic\nGaussian kernels; the downsampling is randomly chosen from nearest, bilinear\nand bicubic interpolations; the noise is synthesized by adding Gaussian noise\nwith different noise levels, adopting JPEG compression with different quality\nfactors, and generating processed camera sensor noise via reverse-forward\ncamera image signal processing (ISP) pipeline model and RAW image noise model.\nTo verify the effectiveness of the new degradation model, we have trained a\ndeep blind ESRGAN super-resolver and then applied it to super-resolve both\nsynthetic and real images with diverse degradations. The experimental results\ndemonstrate that the new degradation model can help to significantly improve\nthe practicability of deep super-resolvers, thus providing a powerful\nalternative solution for real SISR applications.\n