First steps toward image phylogeny

Abstract

Abstract-In this paper, we introduce and formally define a new problem, Image Phylogeny Tree (IPT): to find the structure of transformations, and their parameters, that generate a given set of near duplicate images. This problem has direct applications in security, forensics, and copyright enforcement. We devise a method for calculating an asymmetric dissimilarity matrix from a set of near duplicate images. We also describe a new algorithm to build an IPT. We also analyze our algorithm's computational complexity. Finally, we perform experiments that show nearperfect reconstructed IPT results when using an appropriate dissimilarity function.

Video Phylogeny: Recovering near-duplicate video relationships

“…Regardless of the general philosophy, in the past decade we have seen some progress toward the development of effective systems to identify the cohabiting versions of images [1], [2] and videos [3], [4] in the wild. However, only recently there were the first attempts to go beyond NDDR with techniques to identify the structure of relationships within a set of near-duplicates [5]- [7].…”

“…Different from the Kennedy et al's approach [7], recently two research groups have concurrently focused on the discovery of the structure of dependencies within a set of images and on the underlying conditions of such dependencies [5], [6].…”

“…Only recently there were initial attempts to identify the structure of relationships within a set of near-duplicates [5]- [7], however, limited to the context of images. In this paper, we are interested in the history of the transformations that generated a given set of near-duplicate videos.…”

“…In this paper, we extend the method for calculating an asymmetric dissimilarity matrix from a set of near-duplicate images we introduced in [5] and introduce five approaches to find the phylogenetic relationships of the videos based on the calculated dissimilarities. The first four approaches calculate the final video phylogeny tree based on operations performed directly over the dissimilarity matrices obtained with the video frames.…”

“…The fifth approach calculates the final video phylogeny tree through operations performed over the trees obtained with the video frames. To validate the approaches, we conduct experiments with Youtube videos and use the evaluation methodology introduced in [5].…”