A Study on the Implementation of High Definition GAN Using Multi-Magnification Upscaling

This study proposes a novel super-resolution approach using gradual real number multiplication upscaling to improve image quality beyond conventional integer-based upscaling methods. Unlike standard EDSR (Enhanced Deep Super-Resolution), which performs high-magnification upscaling in a single step, the proposed real_edsrx4 (Progressive 4x Upscaling Model) model progressively applies 1.4× upscaling multiple times before reaching the final resolution. This method preserves high-frequency details more effectively while reducing distortions and blurring. Experimental results on benchmark datasets, including DIV2K and BSD100, confirm that EDSR-P4 outperforms conventional methods in both PSNR and SSIM, producing clearer and more natural images. These findings demonstrate the effectiveness of gradual upscaling in super-resolution tasks, suggesting its potential for high-quality image restoration.