University of Marburg at TRECVID 2006: Shot Boundary Detection and Rushes Task Results

摘要: In this paper, we summarize our results for the shot boundary detection task and rushes at TRECVID 2006. The approach which was evaluated last year 2005 served as a basis experiments modified in several ways. First, investigated different parameter settings unsupervised approach. Second, experimented with possibility to create an ensemble that consists of clusterings have been obtained settings. Our prototype summarization component retrieval component. Rushes videos are segmented on sub-shot order separate redundant from non-redundant sequences within shot. is based subshots clustering, appropriate visualization clusters presented user. sub-shots clustered respect number low-level mid-level features, they visualized such user can navigate through these sub-shots. enables search material automatically according features: camera motion, audio features (silence, speech, music, action, background), speaker identity interviews, sizes, face appearances, by queries example color texture features.