Information Technology Reference
In-Depth Information
moving picture experts group (MPEG-7) visual features and their similarity
matching methods are introduced. Second, experimental evaluation of some
basic visual features is presented to help readers with practical knowledge
of their performance in visual information indexing and retrieval. Then, two
key problems of visual similarity and feature aggregation are discussed for
construction of comprehensive indexing and retrieval systems. The system
performances resulted from different aggregation strategies are demon-
strated. Finally, a practical content-based image retrieval system, Web Image
Search, is presented with its structure and user process in the application of
techniques are presented in this chapter.
11.1 Introduction
Ubiquitous or pervasive computing defines a new paradigm for the twenty-
first century as a result of ubiquity in computing beyond the mainframe,
personal computers, and advances of the media sensing, computing, and
usage along with the developments and applications of communication and
networking. Ubiquitous multimodal sensors capture visual information in
the form of moving and still pictures. In many applications, visual informa-
tion needs to be organized according to relevance or compared with exam-
ples in a networked database to provide service to users. In either case, it is
necessary to index the visual information based on its perceptual content
and extract multimedia documents.
Scenario 1:
In a ubiquitous environment, often networked with the
Internet, certain images are required to be searched over the Internet.
For example, a user may take a picture of a landmark location and then
search for similar pictures taken by other people that would be available
on the Internet, anywhere and everywhere.
Scenario 2:
All images or videos captured by wearable cameras are kept
in a personal multimedia database called lifelog image/video, which
is ubiquitous since images or videos are taken everywhere. Indexing,
retrieval, and summarization of the lifelong images and videos will help
the user to relate his or her past experience.
Scenario 3:
The ubiquitous videos are captured by surveillance cameras
in multiple locations (e.g., shops, hospitals, streets, etc.). When some inci-
dent occurs, the surveillance camera key video frames could be retrieved
to identify certain activities.
This chapter will describe the key techniques for indexing and searching
for visual information including visual information description, visual simi-
larity, and feature aggregation.
