AdvHaze: Adversarial Haze Attack

TL;DR

AdvHaze introduces a realistic haze-based adversarial attack method that effectively misleads neural network classifiers, demonstrating high success and transferability across models, thus advancing non-noise adversarial techniques.

Contribution

This paper presents a novel haze-based adversarial attack method grounded in atmospheric scattering models, expanding beyond traditional noise perturbations.

Findings

High attack success rate on ImageNet and NIPS 2017 datasets.

Better transferability across different models compared to baselines.

Visualization of correlation matrices suggests combining perturbations improves success.

Abstract

In recent years, adversarial attacks have drawn more attention for their value on evaluating and improving the robustness of machine learning models, especially, neural network models. However, previous attack methods have mainly focused on applying some $l^p$ norm-bounded noise perturbations. In this paper, we instead introduce a novel adversarial attack method based on haze, which is a common phenomenon in real-world scenery. Our method can synthesize potentially adversarial haze into an image based on the atmospheric scattering model with high realisticity and mislead classifiers to predict an incorrect class. We launch experiments on two popular datasets, i.e., ImageNet and NIPS~2017. We demonstrate that the proposed method achieves a high success rate, and holds better transferability across different classification models than the baselines. We also visualize the correlation matrices, which inspire us to jointly apply different perturbations to improve the success rate of the attack. We hope this work can boost the development of non-noise-based adversarial attacks and help evaluate and improve the robustness of DNNs.