THE PROBLEM OF IMAGE SUPER-RESOLUTION, DENOISING AND IMAGE RESTORATION METHODS IN DEEP LEARNING MODELS

Abstract

This article addresses the challenges of image super-resolution and noise reduction, which are crucial for enhancing the quality of images derived from low-resolution or noisy data. Several approaches were compared and assessed for upgrading low-resolution images to higher resolutions and eliminating unwanted noise while maintaining the essential characteristics of the original images and recovering images from poor quality or damaged data using deep learning models. It is indicated by research analysis and the experimental outcomes on image quality metrics that the ED-Unet neural network model, enhanced with pretrained weights, significantly outperforms other methods, achieving a Train PSNR of 30.791, a Valid PSNR of 30.699, and a Test PSNR of 31.0172.