Toward an Adaptive Video Retrieval System
Frank Hopfgartner, Joemon M. Jose · Auerbach Publications eBooks · 2009
Unlike text retrieval systems, retrieval of digital video libraries is facing a challenging problem: the semantic gap. Th is is the diff erence between the low-level data representation of videos and the higher level concepts that a user associates with video. In 2005, the panel members of the International Workshop on Multimedia Information Retrieval identifi ed this gap as one of the main technical problems in multimedia retrieval (Jaimes et al. 2005), carrying the potential to dominate the research eff orts in multimedia retrieval for the next few years. Retrievable information such as textual sources of video clips (i.e., speech transcripts) is often not reliable enough to describe the actual content of a clip. Moreover, the approach of using visual features and automatically detecting high-level concepts, which have been the main focus of study within the international video processing and evaluation campaign TRECVID (Smeaton et al. 2006), turned out to be insuffi cient to bridge the semantic gap.