An Introduction to the Principles and Requirements of Robust Hashing

Introduction

A robust hash is essentially a low-dimensional representation of a multimedia signal which is linked to its perceptual content. This representation acts as a descriptor which allows for flexible identification of this type of signal. Flexibility is needed because any multimedia signal —that is, image, audio, or video— may appear in slightly different but perceptually equivalent forms owing to lossy compression, transcoding, clipping/cropping, or a number of other reasons. Therefore, identifiers strictly associated to a particular bit representation of a multimedia signal —such as a cryptographic hash [1]— are far too strict. If robust hashing is to become something like “cryptographic hashing for multimedia” then it is reasonable to expect from it key properties of cryptographic hashing while simultaneously trying to overcome its intrinsic limitation with respect to multimedia signals. Exploiting source coding with distortion constraints is the natural way to achieve this end. The connection between these two areas is explored further on in this chapter in an attempt to produce an accurate definition of robust hashing and its requirements conforming to the point of view expressed above. This point of view is perhaps not the most common in the literature of the subject, which has mainly dealt with the production of robust low-dimensional descriptors but not so much with their connection with cryptographic hashing.