Similarity Preserving Hashing: Eligible Properties and a New Algorithm MRSH-v2

Abstract

Hash functions are a widespread class of functions in computer science and used in several applications, e.g. in computer forensics to identify known files. One basic property of cryptographic Hash Functions is the avalanche effect that causes a significantly different output if an input is changed slightly. As some applications also need to identify similar files (e.g. spam/virus detection) this raised the need for Similarity Preserving Hashing. In recent years, several approaches came up, all with different namings, properties, strengths and weaknesses which is due to a missing definition.

Based on the properties and use cases of traditional Hash Functions this paper discusses a uniform naming and properties which is a first step towards a suitable definition of Similarity Preserving Hashing. Additionally, we extend the algorithm MRSH for Similarity Preserving Hashing to its successor MRSH-v2, which has three specialties. First, it fulfills all our proposed defining properties, second, it outperforms existing approaches especially with respect to run time performance and third it has two detections modes. The regular mode of MRSH-v2 is used to identify similar files whereas the f-mode is optimal for fragment detection, i.e. to identify similar parts of a file.