Perceptual image super resolution using deep learning and super resolution convolution neural networks (SRCNN)

Abstract

The main objective of Perceptual Image Super Resolution is to obtain a high resoluted image from a normal low resolution image. The task is very simple that we just want to make a Low firmness appearance into a extraordinary resolution image. To perform this task we have various methods like Classical Approach in which we try to maximize the mean squared error, evaluate by PSNR(Peak-Signal-to-Noise-Ratio). The first method used to perform this operation was SRCNN (Super Resolution Convolution Neural Network) and these days many of them use DRCN and VDSR which are slightly upgraded methods. Another technique used for the purpose of upscaling to get a high resoluted image from normal little resolution image is the state of art by PSNR. This method was a quite simple one in which we take a low determination image as input and place in a convolution neural network(CNN) and produce a high resolution image as the output. In this technique the edges will be clearly defined, but the whole image will be blurred. This method is unable to produce good-looking textures.