The Human Visual System and Adversarial AI
This work addresses the challenge of making adversarial attacks more perceptually accurate for AI security and robustness applications, but it is incremental as it builds on existing HVS models from image processing.
The paper tackles the problem of approximating perceptual distances in adversarial AI by moving beyond Lp norms to incorporate Human Visual System (HVS) models, demonstrating a proof of concept for improved effectiveness.
This paper applies theories about the Human Visual System to make Adversarial AI more effective. To date, Adversarial AI has modeled perceptual distances between clean and adversarial examples of images using Lp norms. These norms have the benefit of simple mathematical description and reasonable effectiveness in approximating perceptual distance. However, in prior decades, other areas of image processing have moved beyond simpler models like Mean Squared Error (MSE) towards more complex models that better approximate the Human Visual System (HVS). We demonstrate a proof of concept of incorporating HVS models into Adversarial AI.