KDDI LABS at TRECVID 2011: Content-Based Copy Detection.
Yusuke Uchida, Koichi Takagi, Shigeyuki Sakazawa · 2011
We describe our systems for a content-based copy detection (CBCD) task submitted to TRECVID 2011. In this year, focusing on non-geometric transformations, we use only a global visual feature for efficiency. This paper, we describe a fast, accurate content-based video copy detection scheme based on bag-of-global visual features, which is characterized by (1) utilizing an efficient DCT-sign-based feature to enhance fast detection; (2) performing multiple assignment in the temporal domain in addition to the feature and spatial domain, to ensure repeatability in segment-level matching; and (3) adopting inverse document frequency weighting and temporal burstiness-aware scoring to emphasize distinctive visual words. The baseline system processes queries 60 times faster than real-time. The system integrating four baseline systems processes queries 20 times faster than real-time, and it achieves a false negative rate of 1.5 % against transformations 3 and 5 without any false positives. 1.