Title: NAPPure: Enhancing Robustness in Image Classification Against Non-Additive Perturbations via Adversarial Purification

Original: arXiv:2510.14025v2 Announce Type: replace Abstract: Adversarial purification has achieved great success in combating adversarial image perturbations, which are usually assumed to be additive. However, non-additive adversarial perturbations such as blur, occlusion, and distortion are also common in the real world. Under such perturbations, existing adversarial purification methods are much less effective since they are designed to fit the additive nature. In this paper, we propose an extended adversarial purification framework named NAPPure, which can further handle non-additive perturbations. Specifically, we first establish the generation process of an adversarial image, and then disentangle the underlying clean image and perturbation parameters through likelihood maximization. Experiments on GTSRB and CIFAR-10 datasets show that NAPPure significantly boosts the robustness of image classification models against non-additive perturbations.

Rewrite: ArXiv:2510.14025v2 Announcement Type: Replacement

Abstract: While adversarial purification has proven highly effective against additive image perturbations, it often falls short when facing non-additive disturbances like blur, occlusion, and distortion, which are prevalent in practical scenarios. Current purification techniques, optimized for additive noise, exhibit significantly reduced efficacy under these conditions. To address this limitation, we introduce NAPPure, an advanced adversarial purification framework capable of managing non-additive perturbations. Our approach begins by modeling the generation process of adversarial images, subsequently separating the latent clean image from perturbation parameters via likelihood maximization. Empirical evaluations conducted on the GTSRB and CIFAR-10 datasets demonstrate that NAPPure substantially enhances the robustness of image classification models against non-additive threats.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC