Databases Reference
In-Depth Information
Although magnetic disks have been getting higher density as technology in
storage devices rapidly advances, we must depend on data compression for several
or up to ten years in order to construct video databases.
Raw video data, that is to say video data before compression, can be considered
as an ordered set of still images so called frames. We can therefore extract arbitrary
frame from the them. On the other hand, frames in compressed videos, such as
MPEG-1 and MPEG-2 data, are no longer independent, and we cannot simply
extract them. The minimum unit in which we can extract arbitrary frame is called
GOPs (group of pictures), and is about 1/2 sec fragment in case of MPEG-1. We
have constructed a model for compressed videos by taking the above mentioned
characteristics of MPEG-1 into account, and have implemented a video database
system which can extract arbitrary frame based on this model. We also have
employed MPEG-1 as the video format for Heijo. But the operation completely
depends on the facility of ActiveMovie, and thus the minimum unit in which we
can direct is one second.
Users of Heijo can capture a video, and compress it as MPEG-1 data in real
time using a Web browser (Internet Explorer).
In the current system, we must create all indexes by hand. We are planning to
develop automatic generation of indexes in the next step. However, we have
implemented a system to facilitate indexing.
6
Conclusions
One of the most important challenge of video databases is how to recognize
meaningful segments of video data. In this chapter, we have discussed the way to
capture meaningful video data using simple indices. To this end, we have proposed
the “joint” operation, and two methods for utilizing this operation. Finally, we
have reported our prototype video database system “Heijo.”
We have just examined our system by an example of satellite video conference,
which is relatively simple broadcast type video data. For this reason, it is necessary
for us to evaluate our system by more complex examples. Moreover, future work
includes: (1) evaluation of expressive power of our operations, (2) stricter
formalization and modeling, and (3) establishment of operations necessary and
sufficient for acquisition of video objects.
Bibliography
1)
J.F.Allen. Maintaining knowledge about temporal intervals.
Commun. ACM,
26(11):832-843, Nov. 1983.
2)
T.Amagasa, M.Aritsugi, Y.Kanamori, and Y.Masunaga. Interval-based modeling for
temporal representation and operations.
IEICE Trans. on Info. & Syst.,
E81-D(1):47-
55, Jan. 1998.
