Automatic video archaeology: tracing your online videos

We propose in this paper an automatic video archaeology (AVA) system which is able to trace the manipulation history of online videos. The key of AVA is the construction of a video migration map, which represents the history of how videos have been copied or manipulated among a set of near-duplicates. To construct this map, we designed six kinds of video manipulation detectors, including spatial and temporal scaling detectors, spatial and temporal overlay detectors, boarder detector, and grayscale detector. There are a variety of applications based on the proposed AVA system, such as better understanding of the meaning and context conveyed by the manipulated videos, improving current video search engines by better presentation based on the migration map and better indexing scheme based on the annotation propagation. We evaluated the system over a video set with 12,790 videos and 3,481 duplicates. The experimental results show that AVA is able to effectively discover the manipulation relationship among the near-duplicate videos.