by Claudia Hauff and Robin Aly and Djoerd Hiemstra
In this paper we investigate how a small number of high-level concepts derived for video shots, such as Sports, Face, Indoor, etc., can be used effectively for ad hoc search in video material. We will answer the following questions: 1) Can we automatically construct concept queries from ordinary text queries? 2) What is the best way to combine evidence from single concept detectors into final search results? We evaluated algorithms for automatic concept query formulation usingWordNet based concept extraction, and we evaluated algorithms for fast, on-line combination of concepts. Experimental results on data from the TREC Video 2005 workshop and 25 test users show the following. 1) Automatic query formulation through WordNet based concept extraction can achieve comparable results to user created query concepts and 2) Combination methods that take neighboring shots into account