Artificial Intelligence and Communication Technologies
Low Resolution Image Enhancement Using Res-Net GAN
Authors: Arun Sai Narla, Shalini Kapuganti and Hathiram Nenavath
Publishing Date: 20-01-2024
ISBN: 978-81-955020-5-9
Abstract
Though Deeper Convolutional Neural Networks performs better in terms of speed and accuracy of SISR (Single Image Super Resolution), one essential challenge faced is to regain texturing details that are finer while resolving at a greater up-scaling factor. The objective function determines the characteristics of optimization-based super-resolution algorithms. The mean squared reconstruction error has been a major focus of recent research. The resultant approximates had excellent (PSNR) peak signal-to-noise ratio, but they frequently lack great-frequency features and are conceptually unsatisfactory and fall short of fidelity expected at greater resolution. Res-Net GAN is a typical generative adversarial network for super resolution (SR) of image, is presented in the study. Res-Net GAN, a framework which is apt of concluding photo realistic images at a 4x upscaling factors. Here, we present perceptual loss function that comprises of the adversarial loss as well as content loss to achieve this. Using only a discriminator network that was trained to discern among super resolved pictures as well as actual photo realistic pictures, adversarial loss drives solution to natural image manifold. Furthermore, rather than pixel space resemblance, we apply content loss that has been driven by perceptual resemblance. On available standards, deep residual network is capable of recovering photo realistic details from highly down sampled photos. Res-Net GAN exhibits tremendously substantial improvements in imperceptibility in a comprehensive mean-opinion-score (MOS) test. MOS scores achieved with Res-NetGAN were nearer to those achieved with the actual high-resolution pictures.
Keywords
Generative Adversarial Network (GAN), Super Resolution (SR), Mean Opinion Score (MOS)
Cite as
Arun Sai Narla, Shalini Kapuganti and Hathiram Nenavath, "Low Resolution Image Enhancement Using Res-Net GAN", In: Saroj Hiranwal and Garima Mathur (eds), Artificial Intelligence and Communication Technologies, SCRS, India, 2024, pp. 1143-1151. https://doi.org/10.52458/978-81-955020-5-9-108