Most recent single image super resolution (SISR) reconstruction methods adopt simple bicubic downsampling to construct low-resolution (LR) and high-resolution (HR) pairs for training. Those models learn an inverted version of an ideal degradation operation which leads to generating less realistic SR images. The obtained details are either blurred or not reminiscent of the usually observed textures Du et al. ( 2020). The generation of SR image from a single LR with faithful ground-truth texture and no external information remains an issue, especially when the degradation model is not defined (not necessarily bicubic downscaling). To overcome this issue, we focus on designing a single-image SR reconstruction framework for real-world scenarios by injecting the image-specific degradation kernel in the training process. Our method combines the advantages of both SISR and multiple-image super resolution (MISR) techniques by generating a dataset regarding internal statistic of the LR image. A small CNN is trained over this internal dataset and requires no additional or external data. Our method is proved to address more textural details in the generated outcome, and outperforms state-of-the-art deep models.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.