Image phylogeny through dissimilarity metrics fusion

Nowadays, multimedia objects can be easily modified, shared, and distributed, thus determining the widespread diffusion of multiple near-duplicate versions, i.e., objects obtained applying a set of processing operations to original content. This is the case of images downloaded from sharing platforms, modified (e.g., by performing color correction, splicing, etc.) and re-distributed. The evolution of a group of near-duplicate images (i.e., their phylogeny) is a powerful clue to determine both image authenticity and its origin. For this reason, the forensics community has proposed a set of possible solutions to perform phylogenetic analyses based on image dissimilarity computation. Here, we compare different image dissimilarity metrics, and propose a set of original strategies for image phylogeny tree reconstruction. The validation of the proposed methods is performed on a dataset of image phylogeny trees. Depending on the used evaluation metrics, some approaches are preferable to others according to the results. Hence, an analyst can choose the appropriate method according to its needs.