An exact image hash collision between asource and target image can be produced via minuscule adversarial perturbations . In a white box setting, these collisions can be replicated across nearly every image pair and hash type . By attacking points other than the output of a hashing function, an attacker can avoid having to know the details of aparticular algorithm, resulting in collisions that transfer across differenthash sizes or model architectures . Using these techniques, an adversary canpoison the image lookup table of a duplicate image detection service, resultingin undefined or unwanted behavior . Finally, we offer several potentialmitigations to gradient-based image hash attacks to stop these attacks . We demonstrate that these techniques can be used to attack the image .

Author(s) : Brian Dolhansky, Cristian Canton Ferrer

Links : PDF - Abstract

Code :

Keywords : image - attacks - hash - collision - techniques -

Leave a Reply

Your email address will not be published. Required fields are marked *