P‐28: Advanced Image Comparison Metric for Discerning Subtle Distinctions in Visual Quality

Traditional image quality evaluation methods often fail to detect subtle differences crucial for high‐stakes applications. Until now, discerning these minute variations has largely been a manual endeavour that is expensive, time‐consuming, and lacks repeatability, while existing algorithms like PSNR, SSIM, LPIPS, and FID lack sensitivity. This paper introduces three advancements: effective use of existing metrics for subtle comparisons, a new no‐reference metric for detecting fine artifacts, and a novel methodology for detailed image comparison. Using image super‐resolution (SR) as a case study, our method captures critical nuances missed by previous analyses.