Known-Item Search by MCG-ICT-CAS
暂无分享,去创建一个
This paper describes the highlights of known-item search system for TRECVID 2010. We first propose that there lies Understanding Gap between a video’s author and user, which gap has been represented in the author labeled semantic text(sText) description and user generated visual text(vText) query. To bridge this gap, we explore the structured online knowledge from Wikipedia and the data-driven statistics from Google search engine to build map between sText and vText. The experiment results in this KIS task is promising. Meanwhile, by exploring all kinds of visual based methods, we conclude that the great diversity of the web video’s content has made it difficult to find effective visual reference materials from Web, which leads to pool results for visual based methods.
[1] Yongdong Zhang,et al. Tag transformer , 2010, ACM Multimedia.
[2] Yongdong Zhang,et al. Context-oriented web video tag recommendation , 2010, WWW '10.
[3] Paul Over,et al. Evaluation campaigns and TRECVid , 2006, MIR '06.
[4] Takahiro Hara,et al. Wikipedia Mining for an Association Web Thesaurus Construction , 2007, WISE.
[5] Chong-Wah Ngo,et al. Towards optimal bag-of-features for object categorization and semantic video retrieval , 2007, CIVR '07.