Deep Learning for Single Image Super-Resolution: A Brief Review

Abstract

Single image super-resolution (SISR) is a notoriously challenging ill-posed\nproblem, which aims to obtain a high-resolution (HR) output from one of its\nlow-resolution (LR) versions. To solve the SISR problem, recently powerful deep\nlearning algorithms have been employed and achieved the state-of-the-art\nperformance. In this survey, we review representative deep learning-based SISR\nmethods, and group them into two categories according to their major\ncontributions to two essential aspects of SISR: the exploration of efficient\nneural network architectures for SISR, and the development of effective\noptimization objectives for deep SISR learning. For each category, a baseline\nis firstly established and several critical limitations of the baseline are\nsummarized. Then representative works on overcoming these limitations are\npresented based on their original contents as well as our critical\nunderstandings and analyses, and relevant comparisons are conducted from a\nvariety of perspectives. Finally we conclude this review with some vital\ncurrent challenges and future trends in SISR leveraging deep learning\nalgorithms.\n

Keywords

Affiliated Institutions

• University of Rochester US
• Tsinghua University CN
• University Town of Shenzhen CN
• University College London GB

Related Publications