Video alignment for phylogenetic analysis

The possibility of studying multiple objects at once for forensic analysis has paved the way to the development of multimedia phylogeny algorithms. Concerning video phylogeny, a fundamental step at the base of many applications is multiple video alignment. This is, given a pool of near-duplicate video sequences partially overlapping in the temporal domain, find the relative time delay between all of them. As phylogeny methods typically takes into account huge quantities of data, the used alignment algorithms must be computationally efficient. In this paper, we propose a solution for multiple video alignment based on the minimisation of a least-square cost function. The proposed solution can be computed in closed form with reduced computational complexity. Moreover, we propose two possible solutions for refining the estimated alignment based on the removal of outlier measurements.