Combining multiple representations on the TRECVID search task
Arjen P. de Vries, Thijs Westerveld, T.I. Ianeva · 2004
This paper1 presents a (preliminary) analysis of the evaluation re-sults obtained on the TRECVID 2003 search task. We study in particular the effects of combining multiple representations on re-trieval: multiple representations of video content (speech and vi-sual) and of the user information need (multiple visual examples). We conclude from our multi-modal retrieval experiments the fol-lowing working hypothesis: even though the ASR run is usually better than the visual run, matching against both modalities en-sures robustness against choosing the wrong content representa-tion. For the same reason, using multiple visual examples to rep-resent the user information need is preferable over using a single designated example only. 1.