The Human Visual System and Adversarial AI

TL;DR

This paper explores integrating Human Visual System models into Adversarial AI to improve its effectiveness by moving beyond traditional Lp norm measures and adopting more perceptually accurate models.

Contribution

It introduces a proof of concept for incorporating Human Visual System models into Adversarial AI, advancing beyond simple perceptual distance metrics.

Findings

HVS-based models can better approximate human perception of images.

Incorporating HVS models improves adversarial example generation.

Traditional Lp norms are less aligned with human perception.

Abstract

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.