文档介绍：3
Pattern Recognition for Multimedia Content
Analysis
Elena Ranguelova1 and Mark Huiskes2
1 Centrum voor Wiskunde en Informatica (CWI)
2 Universiteit Leiden, LIACS
Introduction
Recognizing Patterns in Multimedia Content
This chapter looks at the basics of recognizing patterns in multimedia con-
tent. Our aim is twofold: first, to give an introduction to some of the general
principles behind the various methods of pattern recognition, and second, to
show what role these methods play in multimedia content analysis.
We start by diving right in by exploring two examples that give a first
flavor of how pattern recognition can contribute to a better understanding
and description of multimedia content.
Example 1: Semi-automatic Annotation of Multimedia Content
One of the foremost uses of pattern recognition is in fulfilling the need for
high-quality metadata, a key ingredient of essful multimedia retrieval sys-
tems. Low-level multimedia bitstreams are not suitable for searching directly,
and pattern recognition is needed to obtain more meaningful and useful de-
scriptions of the data.
Traditionally, multimedia retrieval systems have been based on manual
annotations of the content. Given the typical quantities of material produced,
such annotation is generally a labor-intensive, and thus also expensive, task.
One may think, for example, of the BBC pany which, on a
daily basis, needs to archive material of four television stations as well as a
large number of radio stations. Additionally, for some programs, a single hour
of broadcasting may require an archivist more than 7 hours of cataloging.
For many applications such manual annotation is not very practical, and
thus we have a natural need to automate this process. And although it is
currently not yet possible to design systems that annotate with the same level
of detail as that of well-trained human annotator, there is nevertheless great
scope for systems that provide some basic a