{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/image-super-resolution-via-deterministic","title":"Image Super-Resolution via Deterministic-Stochastic Synthesis and Local Statistical Rectification","arxiv_id":"1809.06557","date":"2018-09-18","proceeding":null,"authors":["Weifeng Ge","Bingchen Gong","Yizhou Yu"],"abstract":"Single image superresolution has been a popular research topic in the last\ntwo decades and has recently received a new wave of interest due to deep neural\nnetworks. In this paper, we approach this problem from a different perspective.\nWith respect to a downsampled low resolution image, we model a high resolution\nimage as a combination of two components, a deterministic component and a\nstochastic component. The deterministic component can be recovered from the\nlow-frequency signals in the downsampled image. The stochastic component, on\nthe other hand, contains the signals that have little correlation with the low\nresolution image. We adopt two complementary methods for generating these two\ncomponents. While generative adversarial networks are used for the stochastic\ncomponent, deterministic component reconstruction is formulated as a regression\nproblem solved using deep neural networks. Since the deterministic component\nexhibits clearer local orientations, we design novel loss functions tailored\nfor such properties for training the deep regression network. These two methods\nare first applied to the entire input image to produce two distinct\nhigh-resolution images. Afterwards, these two images are fused together using\nanother deep neural network that also performs local statistical rectification,\nwhich tries to make the local statistics of the fused image match the same\nlocal statistics of the groundtruth image. Quantitative results and a user\nstudy indicate that the proposed method outperforms existing state-of-the-art\nalgorithms with a clear margin.","url_abs":"http://arxiv.org/abs/1809.06557v1","url_pdf":"http://arxiv.org/pdf/1809.06557v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"image-super-resolution-via-deterministic","repo_url":"https://github.com/Wenri/DTSN","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"image-super-resolution","task_name":"Image Super-Resolution"},{"task_slug":"super-resolution","task_name":"Super-Resolution"},{"task_slug":"regression-1","task_name":"regression"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1809.06557","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}