Fischlar-TRECVid-2004: combined text- and image-based searching of video archives

N.A. O'Connor, Hyowon Lee, A.F. Smeaton, G.J.F. Jones, E. Cooke, H. Le Borgne, C. Gurrin
2006 IEEE International Symposium on Circuits and Systems  
The Fischlar-TRECVid-2004 system was developed by typing some text and/or adding video/image examples that for Dublin City University's participation in the 2004 TRECVid come with the topic. During the search, the user can include video information retrieval benchmarking activity. The system any keyframes determined as relevant from the video into allows search and retrieval of video shots from over 60 hours of content. The shot retrieval engine employed is based on a subsequent queries. System
more » ... ent queries. System B relied solely on searching based combination of query text matched against spoken dialogue com-on keyframe images without any text-based querying. Thus bined with image-image matching where a still image (sourced the only way the user can initiate a search in System B is externally), or a keyframe (from within the video archive itself), by including video/image examples in a search query panel. is matched against all keyframes in the video archive. Three Thereafter, keyframes can be added to or removed from the separate text retrieval engines are employed for closed caption . ' text, automatic speech recognition and video OCR. Visual shot quer in anate t to improve the search result. T he system matching is primarily based on MPEG-7 low-level descriptors. was used to conduct an interactive search experiment for the The system supports relevance feedback at the shot level enabling 25 topics provided for the 2004 TRECVid interactive search augmentation and refinement using relevant shots located by task. For our experiments, 16 experienced users carried out the user. Two variants of the system were developed, one this search task each working under a time constraint of 15 that supports both text-and image-based searching and one that supports image only search. A user evaluation experiment minutes per topic. compared the use of the two systems. Results show that while The remainder of this paper is organised as follows. Secthe system combining text-and image-based searching achieves tion II presents an overview of the system, including degreater retrieval effectiveness, users make more varied and scriptions of the text and image search engines and how the extensive queries with the image only based searching version. results of these are combined (sections II-A, II-B and II-D, respectively). The relevance feedback approach employed
doi:10.1109/iscas.2006.1693029 dblp:conf/iscas/OConnorLSJCBG06 fatcat:qofvytedpvb3zl5bnhsmhntyji