메인 내용으로 이동

Perceptual Hashing

Intentionally maximizes the change of hash collision to detect similar images. Used by Google's Search by Image feature or Apple's CSAM. Perceptual Hashing

Algorithms and Their Performances


  • Fastest: block-mean-based perceptual image hash.
  • Most discriminative: Marr-Hildreth operator-based hash function


It's Not What It Looks Like: Manipulating Perceptual Hashing-based Applications

  • TLDR: A delicately designed image can fool a perceptual hash application, making it classify two completely different images as the same.