Analyzing Adversarial Robustness of Deep Neural Networks in Pixel Space: a Semantic Perspective

TL;DR

This paper investigates the semantic-based pixel-level vulnerabilities of deep neural networks to adversarial attacks, revealing that only specific regions with certain semantic content are critical for successful perturbations.

Contribution

It introduces a novel algorithm to identify vulnerable pixels in different semantic regions, highlighting the importance of semantic information in adversarial robustness.

Findings

Successful one-pixel attacks target specific semantic regions.

Vulnerable points are scattered across different image regions.

Robustness varies with the semantic content of image regions.

Abstract

The vulnerability of deep neural networks to adversarial examples, which are crafted maliciously by modifying the inputs with imperceptible perturbations to misled the network produce incorrect outputs, reveals the lack of robustness and poses security concerns. Previous works study the adversarial robustness of image classifiers on image level and use all the pixel information in an image indiscriminately, lacking of exploration of regions with different semantic meanings in the pixel space of an image. In this work, we fill this gap and explore the pixel space of the adversarial image by proposing an algorithm to looking for possible perturbations pixel by pixel in different regions of the segmented image. The extensive experimental results on CIFAR-10 and ImageNet verify that searching for the modified pixel in only some pixels of an image can successfully launch the one-pixel adversarial attacks without requiring all the pixels of the entire image, and there exist multiple vulnerable points scattered in different regions of an image. We also demonstrate that the adversarial robustness of different regions on the image varies with the amount of semantic information contained.