Generative Adversarial Network-Based Methods in Super Resolution

Abstract:
Super-resolution (SR) is a crucial image processing technique to optimize the resolution of images and videos. Recent years have witnessed significant development of SR approaches using Generative Adversarial Nets (GAN). Herein, a thorough overview on the latest achievements of SR approaches using GAN are given. Specifically, we firstly define the SR problem. Then, we introduce traditional and deep-learning (DL)-based SR techniques including Convolutional Neural Network (CNN) and GAN methods. Afterward, we explore the background of GAN and pay special attention to GAN-based approaches such as SRGAN, GMGAN, and Cycle-GAN. In addition, we also cover the applications of GAN-based SR approaches in real life, including medical diagnosis and remote sensing.