Databases Reference
In-Depth Information
achieved with motion pictures only, or with audio media alone. Cooperative
processing of simple indices is required. In addition, handling multiple scenes that
occurred at the same time, but recorded on different videos, cannot be done
effectively with conventional (individual) indexing method. Here, again
“cooperative” processing of indices becomes important.
In this chapter, we presume simple indexing for each single media is available.
We then discuss a method that finds out semantically coherent video objects while
using this indexing in a practical manner. By “video object” we mean “semantically
coherent video information (picture and sound)”. A video object is defined by a
time interval i.e. by a pair of
7)
.
We propose a “joint” operation for time intervals, introduce its applications,
and report an experimental implementation of a prototype video database system
called the “Heijo” system.
Related research works on video database systems include the OVID system
8)
and Video Algebra
9,
10)
. The OVID system uses a video object data model, and
combines multimedia data with an object-oriented database. Video Algebra is the
proposal of a set of operation for carrying out such operations as connection of
videos, parallel and simultaneous “play”, or repeated “play”. They are mainly
oriented toward the playing of video, and database-type operations are never
considered. A well known research on searching for semantically coherent objects
from among a flood of video information is Infor Media. Infor Media, advocated
a search for necessary information by the quoting of all necessary means
3,
4)
.
As for time interval operations, there is the pioneering work by Allen who
classified time-based relations
1)
. Since then, there have been proposals on time
interval relations related to sets of multiple time intervals
5,
6)
, research to introduce
empty time intervals
7,
2)
, and research on handling all multimedia objects in a
unified way through time intervals
7)
Our research starts from a position close to InforMedia, but we discuss an
original method that searches for meaningful video fragments (semantically
coherent video object) by combining simple indices, and propose a new time interval
operation for that purpose. In addition, we take up the problem of operations on
a number of different media over the same time axis, and discuss how we can
apply the proposed video operation even to relations of the multiple media whereby
separate indexing is performed for each.
start time
and
end time
2
Research objective
Suppose we have simple and separate indices for pictures and sound of video
information. Our target is to search for designated video object utilizing those
simple indices.
By indexing, it becomes possible to search for a picture in which a person
identified as Mr. X is reflected or to search for a picture in which Mr. X is talking
(actually, this is a research theme in itself, but here we proceed with the research
