Block-Wise Image Transformation With Secret Key for Adversarially Robust Defense

In this paper, we propose a novel defensive transformation that enables us to maintain high classification accuracy under the use of both clean images and adversarial examples for adversarially robust defense. The proposed is block-wise preprocessing technique with secret key input images. defense obfuscates gradients in absence unlike previously defeated obfuscating defenses. We developed three algorithms realize transformation: Pixel Shuffling, Bit Flipping, FFX Encryption. Experiments were carried out on CIFAR-10 ImageNet datasets by using black-box white-box attacks various metrics including adaptive ones. results show achieves close even first time. best-case scenario, model trained transformed Encryption (block size 4) yielded an 92.30% 91.48% PGD attack noise distance 8/255, which non-robust (95.45%) dataset, it 72.18% 71.43% same attack, also standard (73.70%) dataset. Overall, all are demonstrated outperform state-of-the-art defenses training whether or not attack.