This paper investigates the issue of using a combination of motion descriptors, computed directly from MPEG motion vectors, for effective dynamic video content analysis and characterisation. On a shot by shot basis, the descriptors describe the general motion activities as well as the spatial distribution of motions within the sjot. On a frame by frame basis, they represent the continuous changes in pace and dynamics of the underlying video content. The formed lead itself to an efficient motion-based shot retrieval scheme while being supported by a simple but effective lateral ranking fusion technique. It can be used for story segmentation and video summarisation too. And the latter is suitable for video skimming and fast video browsing. The system can be easily integrated with existing video retrieval and story segmentation system using only static visual features.
Using Lateral Ranking for Motion-Based Video Shot Retrieval and Dynamic Content Characterization
2005-01-01
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
DDC: | 629 |
British Library Online Contents | 2004
|Fuzzy framework for unsupervised video content characterization and shot classification
British Library Online Contents | 2001
|Content-based retrieval from digital video
British Library Online Contents | 1999
|A Bayesian Video Modeling Framework for Shot Segmentation and Content Characterization
British Library Conference Proceedings | 1997
|