An Unsupervised Anchorperson Shot Detection Based on the Distribution Properties

Anchorperson shot detection is the vital step to parse news video and index news key information. However, most current methods for anchorperson detection mainly depend on temple matching, which inevitably renders limitation to diversified styles of news programs. In this paper, we propose a novel model-free anchorperson detection algorithm based on anchorperson distributing traits. First, news video is divided into shots, and then candidate shots are selected according to one of the anchorperson attribute. Second, cluster algorithm is applied to the candidate shots for gathering the vision similar shots. Finally, the variance analysis on basis of another anchorperson attribute is applied to each cluster for eventually determining the anchorperson shots. Experiments results show that our method has achieved excellent results in various kinds of news video.