The paper presents a method to detect identical video segments from video footage in broadcast video archives, and its application to news video analysis. We define identical video segments as distinctively similar video segments. After giving the definition of identical images, an efficient algorithm to detect identical shots in videos is presented. Its effectiveness is shown by applying the algorithm to news video analysis. As the first experiment, the algorithm successfully locates identical shots derived from the same video materials shared by semantically related topics. This obviously is useful for topic tracking by visual cues. The second experiment shows that aggregating detected identical shots can tell the structure of the news video, i.e., can locate filler shots such as opening shots, anchor shots, weather forecast, etc. The result is quite helpful for news video parsing. The experiments reveal that, although the method does not use any news-specific a priori knowledge, it could be used as a powerful tool to explore useful knowledge from large-scale news video archives.
[1]
Shin'ichi Satoh,et al.
The SR-tree: an index structure for high-dimensional nearest neighbor queries
,
1997,
SIGMOD '97.
[2]
Robert M. Haralick,et al.
Graph-theoretic clustering for image grouping and retrieval
,
1999,
Proceedings. 1999 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (Cat. No PR00149).
[3]
Avideh Zakhor,et al.
Video similarity detection with video signature clustering
,
2001,
Proceedings 2001 International Conference on Image Processing (Cat. No.01CH37205).
[4]
Shin'ichi Satoh,et al.
Distinctiveness-sensitive nearest-neighbor search for efficient similarity retrieval of multimedia information
,
2001,
Proceedings 17th International Conference on Data Engineering.