ティモシー シェパード, 田中昭二, 井上誠喜
Querying and Indexing Multimedia Data
Abstract:Multimedia Data is heterogenous as compared to data of other kinds, such as financial or business Data． It is complex in nature, often consisting of generalisation-specialization abstractions and whole-part structure. For example, a video sequence is simply an image with a time dimension(inheritance),
and a movie is a collection of parts, such as video, sound, and script(whole-part).
One interesting problem when dealing with Multimedia databases is to develop a manner in which to query them efficiently, a method which can deal with the complex nature of such data. In this report I will outline a graphical query method for complex multimedia databases.
Another interesting problem, is indexing multimedia data. This can be approached from two perspectives, indexing for the computer, or indexing from the human. For example, a number ID for students is for the computer, while using the name as an ID is for the human. Often, a combination of both is the ideal indexing method for data.
In this report I will detail indexing methods for two of the most important types of multimedia: video and audio. I will describe a method for extracting representative frames for movies and an algorithm for extracting representative samples from instrumental music. The approach is from the human perspective, as the concern here is mostly to represent larger sets of information by key elements for the human user.
Finally I will conclude with some discussion about how the various results of this technical report. can be both integrated into a larger system or used seperately in other applications.