Image super-resolution is a critical aspect of image enhancement, facilitating the reconstruction of high-quality images from low-resolution inputs. Traditional quality assessment metrics like SSIM, MSE, and PSNR have limitations in effectively evaluating super-resolution models due to their focus on pixel values and statistical properties, overlooking overall visual quality. This repository introduces a technique for comparing super-resolution models using a pattern-based approach. It evaluates image quality by analyzing the harmonics, providing a performance comparison index that surpasses traditional metrics. By focusing on the frequency domain and magnitudes of Fourier components, this technique effectively captures image features and patterns, enabling a more comprehensive assessment of super-resolution model performance.
The paper is available to read: https://doi.org/10.1007/s11760-024-03430-8
@article{Koçmarlı2024,
author={Ko{\c{c}}marl{\i}, G{\"o}khan and Esmer, G{\"o}khan Bora},
title={Performance comparison index for image super-resolution models},
journal={Signal, Image and Video Processing},
year={2024},
month={Aug},
day={02},
abstract={Image super-resolution is a critical aspect of image enhancement, facilitating the reconstruction of high-quality images from low-resolution inputs. Traditional quality assessment metrics like SSIM, MSE, and PSNR have limitations in effectively evaluating super-resolution models due to their focus on pixel values and statistical properties, overlooking overall visual quality. This article introduces a technique for comparing super-resolution models using a pattern-based approach. The proposed method evaluates image quality by analyzing the harmonics, providing a performance comparison index that surpasses traditional metrics. By focusing on the frequency domain and magnitudes of Fourier components, this technique effectively captures image features and patterns, enabling a more comprehensive assessment of super-resolution model performance.},
issn={1863-1711},
doi={10.1007/s11760-024-03430-8},
url={https://doi.org/10.1007/s11760-024-03430-8}
}