Each row displays, from left to right: the original (clean) license plate image, the detected license plate region, I-FGSM with EOT (before printing and recapturing), I-FGSM with EOT (after printing and recapturing), I-FGSM with EOT + P&S (before printing and recapturing), and I-FGSM with EOT + P&S (after printing and recapturing). In the first two rows, the fourth and sixth columns demonstrate successful attacks, where the adversarial examples remain effective after the physical transformation, leading to mis-classification by the LPD model. In contrast, the third row shows a failed attack, where the perturbations do not survive the print-and-recapture process in either case. The fourth row highlights the importance of incorporating the P&S simulator within EOT, resulting in a successful misclassification by the LPD model when using I-FGSM with EOT + P&S.