Go Deep into Super-Resolution: A Systematic Review Based on Problem Decomposition and Classification

Abstract

Super-resolution (SR) has been attracting research attention in recent decades because of the growing demand for higher resolution. A large variety of algorithms in the computer vision field was applied to improve this issue. In this paper, we provide a problem-based analysis for the SR problem by dissembling the original issue into several subproblems. We broadly bring these problems under three main levels, information augmentation, mapping, and orientation and control. Furthermore, we elaborate on various research perspectives under each level, summarizing typical and advanced algorithms improved on related processes. Besides, we discuss the similar ideas and evolutionary relationships revealed in different methods. We also arrange some popular datasets and competitions in the SR field. By doing this, we aim to inspire deeper investigation and enlighten possible research sites of SR problems in the future.