Story segmentation of xinwenlianbo based on three-level criterion of content analysis

To detect the story boundary (SB) of Xinwenlianbo precisely, this paper proposes a method of three-level criterion story segmentation (SS) based on content analysis. For the story with anchorperson, this paper uses penny distance to judge whether the anchorperson gotten by video feature is true or not to look for SB; for the story without anchorperson, this paper looks for the SB by judging the shot change point occurs in the silence region. The experiment results show that the method can improve 6.98% precision comparing with only using video feature in anchor detection and the error range can be decreased from 1.5 ~2.5 seconds to 0 ~0.5 second in SB without anchor by silence detection, which settles the problem of no precision boundary and video frame lost corresponding to silence for only using silence .The final precision of SS reaches 93.12%.