Abstract
In this paper we describe our TRECVID 2009 video re- trieval experiments. The MediaMill team participated in three tasks: concept detection, automatic search, and in- teractive search. The starting point for the MediaMill con- cept detection approach is our top-performing bag-of-words system of last year, which uses multiple color descriptors, codebooks with soft-assignment, and kernel-based supervised learning. We improve upon this baseline system by explor- ing two novel research directions. Firstly, we study a multi- modal extension by including 20 audio concepts and fusion using two novel multi-kernel supervised learning methods. Secondly, with the help of recently proposed algorithmic re- nements of bag-of-word representations, a GPU implemen- tation, and compute clusters, we scale-up the amount of vi- sual information analyzed by an order of magnitude, to a total of 1,000,000 i-frames. Our experiments evaluate the merit of these new components, ultimately leading to 64 ro- bust concept detectors for video retrieval. For retrieval, a robust but limited set of concept detectors justi es the need to rely on as many auxiliary information channels as pos- sible. For automatic search we therefore explore how we can learn to rank various information channels simultane- ously to maximize video search results for a given topic. To further improve the video retrieval results, our interactive search experiments investigate the roles of visualizing pre- view results for a certain browse-dimension and relevance feedback mechanisms that learn to solve complex search top- ics by analysis from user browsing behavior. The 2009 edi- tion of the TRECVID benchmark has again been a fruitful participation for the MediaMill team, resulting in the top ranking for both concept detection and interactive search. Again a lot has been learned during this year's TRECVID campaign; we highlight the most important lessons at the end of this paper.