Deep-learning-based single-image super-resolution in practice

The accessibility and affordability of high-fidelity information, such as 4K quality videos and photographs, increase with the availability of high-speed internet connections. Despite advancements in camera sensor technology, artificial enhancements created by smart image processing algorithms significantly enhance image fidelity in recent years. In the field of computer vision, we focus on single-image super-resolution (SISR) technology, which aims to reconstruct high-resolution (HR) images from low-resolution (LR) images, marking it as a critical area of our research. Advances in deep learning theory and technology provide novel solutions for image super-resolution (SR), with outcomes substantially surpassing those of traditional methods in many aspects. In this review article, we concentrate on deep learning-based image super-resolution algorithms, offering a comprehensive introduction from foundational frameworks to datasets. We delve into the application of deep learning in the SISR field by analyzing different foundational framework models. Additionally, we discuss the metrics for evaluating super-resolution effects and compare the performance of various deep learning-based algorithms. We conclude with a forward-looking perspective on the trends in research on deep learning- based image super-resolution algorithms.