AT&T research at TRECVID 2009 content-based copy detection

Abstract

AT&T participated in one evaluation task at TRECVID 2009: the content-based copy detection (CBCD) task. In the proposed CBCD framework, we employed local visual features to match the video content, and adopted locality sensitive hashing techniques to maintain the scalability and the robustness of our approach. This year, we focused on the visual only CBCD task, and submitted three runs: one for NoFA (no false alarm) profile, and two for balanced profile. Evaluation results show that our CBCD system is highly effective and accurate.