Robust features for accurate spatio-temporal registration of video copies

Abstract

Pirate copies of movies are proliferating on the Internet and causing huge piracy issues. Any anti-piracy strategy requires not only copy detection but also precise frame alignment of pirate video with master content, prior to the estimation of geometric distortions and capture location in a theater. Most studies in pirate video registration focus on the alignment of watermarked sequences, while few efforts are made to align non-watermarked videos using content-based features. In this paper, we propose a spatio-temporal scheme for aligning pirate and master contents using visual features, which consists of three stages: First, a video sequence is compactly represented using 1-D SURF (Speeded-Up Robust Features) signatures; Second, temporal frame alignments are computed using sliding window based dynamic programming method; Third, robust SURF descriptors are employed to generate spatial frame alignments. The results demonstrate the improved registration accuracy of the proposed method against various transformations. � 2012 IEEE.