Single Image Super-Resolution

Abstract

Super-Resolution (SR) of a single image is a classic problem in computer vision. The goal of image super-resolution is to produce a high-resolution image from a low-resolution image. This paper presents a popular model, super-resolution convolutional neural network (SRCNN), to solve this problem. This paper also examines an improvement to SRCNN using a methodology known as generative adversarial network (GAN) which is better at adding texture details to the high resolution output.