A robust and low-cost video fingerprint extraction method for copy detection

Video fingerprinting for content-based video identification is a very useful task for the management and monetization of copyrighted content distribution. The main challenges of monitoring and copy detection systems are: a) the effective identification of highly transformed videos (robustness) and b) computational efficiency which may be relevant for some applications. Typically, most video fingerprinting methods focus on robustness leaving aside computational efficiency. However, for real-time applications are necessary low computational cost detection methods, for instance, in illegal content monitoring in video streaming distributions. Therefore, in this paper, we propose a low-cost and effective video fingerprint extraction method based on the combination of content-based features using both acoustic and visual video components. Our method is capable of detecting video copies by using computationally efficient fingerprints while maintaining robustness against the decrease in quality and content preserved distortions, which are frequent but severe attacks.