AAI_2025_Capstone_Chronicles_Combined

15

quality than correctly classified Reals. It may indicate some bias of the model to associate lower image quality with real images and higher image quality with fake ones. Table3 Pearson’s test for relationship between AQI and misclassification rate for CNN PIQE Fake r = 0.313 p<0.001 Real r = -0.270 p<0.001 WADIQAM_NR Fake r = -0.278 p<0.001 Real r = 0.274 p<0.001 The ViT model though only seemed to present a minor significant correlation between quality and falsely classified Fakes for the “TOPIQ_NR” (Chen et al., 2023) algorithm, another neural network based approach. Figure11 Relative misclassification rate compared to “TOPIC_NR” metric for ViT

372