Deep learning methods in real-time image super-resolution: a survey

Abstract

Super-resolution is generally defined as a process to obtain high-resolution images from low-resolution inputs, which has attracted considerable attention from the image-processing community. This paper aims to analyze, compare, and contrast technical problems, methods, and the performance of super-resolution research, especially real-time super-resolution methods based on deep learning structures. Specifically, we first summarize fundamental problems, categorize algorithms, and analyze possible application scenarios. Increasing attention has been drawn to utilizing convolutional neural networks (CNN) or generative adversarial networks (GAN) to predict high-frequency details lost in low-resolution images. We provide a general overview of background technologies and pay special attention to super-resolution methods based on deep learning architectures for real-time processing, which not only produce desirable reconstruction results but also expand the possible applications of super-resolution to systems like cell phones, drones, and embedded systems. Benchmark datasets are enumerated, and the performance of the most representative approaches is compared to provide a fair view of current methods. Finally, we conclude the paper and suggest ways to improve the use of deep learning methods in real-time image super-resolution.

Yirui Wu

Young Professor, CCF Senior Member

My research interests include Computer Vision, Artifical Intelligence, Multimedia Computing and Intelligent Water Conservancy.