Evaluation of a practical video fingerprinting system

Video fingerprinting is one of the most promising technologies to uniquely identify video content in the Internet in a totally format-independent way. The logic behind the creation of the video fingerprints or perceptual hashes is a key factor for the performance of the system, especially in terms of storage and search efficiency. Although several video fingerprinting methods have been proposed, most of them largely focus on the robustness of the perceptual hashes to distortions, giving much less importance to the required storage as well as the efficiency of the database structure for fast and reliable format-agnostic searches of video content. In this paper we propose a video fingerprinting system with an excellent robustness to distortions but also offering lower storage requirements and a more efficient database structure than prior schemes.