Robust Homomorphic Image Hashing

Abstract

Most image forensic techniques are concerned with authenticating the contents of an image, linking an image to a device or class of devices, or extracting forensically useful information from an image. Another aspect of image forensics is the identification of previously identified content, particularly in the face of simple image modifications. Such so-called robust image hashing techniques can be highly effective at finding child sexual abuse material, revenge porn, terrorism-related material, and dangerous or hateful conspiracy material. The growing use of end-to-end encryption on commercial platforms makes identification of such material significantly more challenging. We describe a robust image hashing algorithm that is both robust to simple image manipulations and that can operate on an encrypted image, without the need or even ability to decipher the underlying encrypted image.