Adversarial Color Projection: A Projector-Based Physical Attack to DNNs

Recent advances have shown that deep neural networks (DNNs) are susceptible to adversarial perturbations. Therefore, it is necessary to evaluate the robustness of advanced DNNs using adversarial attacks. However, traditional physical attacks that use stickers as perturbations are more vulnerable than recent light-based physical attacks. In this work, we propose a projector-based physical attack called adversarial color projection (AdvCP), which performs an adversarial attack by manipulating the physical parameters of the projected light. Experiments show the effectiveness of our method in both digital and physical environments. The experimental results demonstrate that the proposed method has excellent attack transferability, which endows AdvCP with effective blackbox attack. We prospect AdvCP threats to future vision-based systems and applications and propose some ideas for light-based physical attacks.