Columbia University's semantic video search engine 2008
Eric Zavesky, Shih‐Fu Chang · 2008
This paper describes the newest revisions of "CuVid", Columbia University's video search engine, a system that enables semantic multimodal search over video broadcast news collections that first evaluated in the NIST TRECVID2005 benchmark and later expanded to include a large number (374) of visual concept detectors. In this work we start with concept classifiers trained with LSCOM (Large Scale Concept Ontology for Multimedia) and propose a new real-time system that interacts with large sets of video data. Extending prior work that identifies dominant concepts from a set of query results, we hope to further remove the requirement of full lexicon knowledge prior to use and introduce instant revision of concept influence as the user explores their result space.