Multi-Clue Image Tampering Localization

Image tampering is nowadays at everyone's reach. This has determined an urgent need of tools capable of revealing such alterations. Unfortunately, while forgeries can be operated in many different ways, forensic tools usually focus on one specific kind of forgeries. Therefore, an effective strategy for tampering detection and localization requires to merge the output of many different forensic tools. In this paper, we propose an algorithm for image tampering localization, based on the fusion of three separate detectors: i) one based on PRNU, working when we have at least a few of pictures shot with the same camera; ii) one based on PatchMatch; iii) one exploiting image phylogeny analysis, in case we have a set of near-duplicate images to analyze. The method is validated against the dataset released by the IEEE Information Forensics and Security Technical Committee for the First Image Forensics Challenge. Results show that the proposed algorithm can beat the challenge with the highest score achieved at paper submission time.