Automatic video search using semantic concepts
This paper describes automatic video search using semantic concept detection, and how it is implemented in our PicSOM system. The demand for such methods is growing because of the exploding growth of digital video data produced today, ranging from professionally produced TV programming to individuals sharing videos online. Content-based methods address the problem of finding relevant information in large amounts of visual data, which by nature are very hard to search and index automatically. The semantic gap between the high-level concepts understood by humans and the low-level information of computer systems is partially bridged by modelling mid-level semantic concepts. The performance of our system is judged in the video retrieval setting of the international TRECVID 2008 and 2009 evaluations, comparing favourably with the competing state-of-the-art systems.