¡@

A video server on the Internet usually provides a short version of a video clip, a video abstraction, in order to provide highlights or the overall story to users as quickly as possible. However, since it is statically generated by content providers using domain-dependent heuristics, it cannot satisfy all users simultaneously. This paper proposes a domain-independent video abstraction algorithm that generates various video abstractions dynamically according to the users¡¦ requirements. It first identifies some low-level visual and temporal constraints that a good video abstraction should satisfy domain-independently, such as that it should be well-distributed, highly-active, and non-duplicated (or concise), that are used to partially represent the user¡¦s requirements. These constraints are formalized as objective functions, and a simulated annealing algorithm is used to find a set of shots that maximizes the weighted sum of these objective functions as much as possible. It is a personalized abstraction algorithm since each user can generate and view various video abstractions by dynamically adjusting the weights of the constraints. From the results of several experiments with a Korean movie and other well-known movie videos, we found that the proposed algorithm can produce various useful video abstractions very quickly.

Keywords: video abstraction, video analysis, video-on-demand, simulated-annealing, video trailer, multimedia system

Retrieve PDF document (200411_09.pdf)

Received December 12, 2002; revised July 2, 2003; accepted Septemer 15, 2003.
Communicated by Liang-Gee Chen. ^*This work was supported by the Basic Research Program of the Korea Science and Engineering Foundation (Grant No. R01-2002-000-00141-0)