Single Picture Super-Resolution Using Generative Adversarial Network

Abstract

Picture Super-resolution is a by and large analysed issue in computer vision, where the objective is to change over a low-resolution picture to a highresolution picture. As of now, deep learning methods such as convolution neural systems and generative adversarial networks are being utilized to perform superresolution with results competitive to the best in class. In this paper, a generative adversarial network, SRGAN, is proposed for super-resolution with a perceptual loss work comprising of an adversarial loss, mean squared loss, and content loss. The target of our usage is to get familiar with an end-to-end mapping between the low and high-resolution pictures furthermore, enhance the up-scaled picture for quantitative measurements as well as absolute quality. We at that point think about our outcomes with the present cutting-edge strategies in super-resolution, lead proof of idea division study to show that super-resolved pictures can be utilized as a compelling pre-processing step before division and approve the findings measurably.