Automatic fast detection of anchorperson shots in temporally aggregated TV news videos

Abstract

The temporal aggregation method applied for sports news videos detects and aggregates two kinds of shots: sequences of long shots, mainly studio shots unsuitable for content-based indexing, and sports player shots adequate for sports categorization. Hereby, it significantly reduces the number of frames analyzed in content-based indexing of TV sports news. The tests have shown that applying the temporal aggregation method it was possible to reject about half of video frames and despite this almost all sports scenes reported in TV sports news have been indexed. The paper examines the influence of the temporal aggregation on the detection of anchorperson shots in news videos. The TV news video editing is similar to that of TV sports news although news shots are longer in average than sports player shots. The interviews, statements, and commentaries are more significant in news than in sports news for contentbased analyses because these statements are not necessarily spoken by an anchorman, so they are usually important informative parts of TV news. The experiments carried out on TV news and described in the paper have shown that anchorperson shots as well as interview shots may be more easily and faster selected when TV news videos are temporally aggregated. These experiments were performed in the Automatic Video Indexer AVI.