Abstract
Many adversarial attacks produce floating-point tensors which are no longer adversarial when converted to raster or JPEG images due to rounding. This paper proposes a method dedicated to quantize adversarial perturbations. This “smart” quantization is conveniently implemented as versatile post-processing. It can be used on top of any white-box attack targeting any model. Its principle is tantamount to a constrained optimization problem aiming to minimize the quantization error while keeping the image adversarial after quantization. A Lagrangian formulation is proposed and an appropriate search of the Lagrangian multiplier enables to increase the success rate. We also add a control mechanism of the $\ell _\infty $ -distortion. Our method operates in both spatial and JPEG domains with little complexity. This study shows that forging adversarial images is not a hard constraint: our quantization does not introduce any extra distortion. Moreover, adversarial images quantized as JPEG also challenge defenses relying on the robustness of neural networks against JPEG compression.
Funding Information
  • Direction Générale de l’Armement, French, under the Ministry for the Armed Forces
  • French National Research Agency under the ALASKA Project (ANR-18-ASTR-0009)
  • DEtection de FALSifications dans des images et vidéos (DEFALS) Program (ANR-16-DEFA-0003)
  • French Agence Innovation pour la Défense under the chaire Security of Artificial Intelligence for Defense Applications (ANR-20-CHIA-0011-01)

This publication has 15 references indexed in Scilit: