ViCopT: a robust system for content-based video copy detection in large databases
This paper presents a novel approach for extracting discriminative descriptions of 3-D objects using spatio-temporal information. In particular, local features are tracked in image sequences leading to local trajectories containing dynamic information. These trajectories are judged with respect to their quality and robustness and finally each of them is assigned a single local descriptor from a key-frame in order to obtain an object description. Extensive experiments compare this novel approach for selecting local features to state-of-the-art view-based methods and show that it outperforms existing methods.