Databases Reference
In-Depth Information
6
Heijo—A Video Database System
for Retrieving Semantically Coherent
Video Information
Shunsuke Uemura
Masatoshi Yoshikawa
Toshiyuki Amagasa
ABSTRACT
This chapter describes a research and development project for a video database system
“Heijo.” Retrieving meaningful video fragments from a video database is a challenging
issue. Heijo can process such a query by combining simple indices constructed on
each video data. To this end, three kinds of operations on time intervals, namely,
intersection, union, and joint operations and algorithms for them are proposed. In
addition, five kinds of time axes of video data and categorization of them based on
the time axes are investigated so that we can distinguish video data by their contents.
Finally, an overview of Heijo system is shown.
1
Introduction
Rapid progress in computer science and network technologies are realizing world
wide information resources. Not only the volumes of those information resources
are increasing, but also the kind of media being digitized is changing, namely from
fairly simple character strings to such entities as those directly appeal to the senses
of humans like pictures and sounds. Next-generation database systems should be
able to integrate and manipulate those digital media.
This chapter describes a research and development project for video database
systems or database systems for video media. We use the terminology “video” as
a generic name for motion picture and its accompanying sound. One of the most
important research topics for video database system is how to capture semantically
unified and meaningful video fragment or semantically coherent video information.
Traditional data model for video consists of “frame (still picture)”, “cut (a series
of frames)”, and “scene (a series of cuts)”, Frame, cut, and scene are all somewhat
physical concepts. It is difficult to get meaningful video fragment from those
concepts. Searching for semantically coherent video information can never be
81
