Image quality constrained GAN for super-resolution

Abstract

As one of the most important research topics in image processing, super-resolution aims to estimate high resolution images from single or multiple low resolution images taken from the same scene. With the advent of deep learning techniques, generative adversarial networks are widely adopted for solving various image processing problems including super resolution. We investigate the effect of introducing image quality constraints into the training objective function of a generative adversarial network for super resolution. Experiment results demonstrate that network performance has great potential to be improved with such constraints.