INTRODUCTION TO CONTENT-BASED VISUAL PROCESSING - Video Object Extraction and Representation: Theory and Applications - page 11

The ability to learn and adapt contextually are strengths of neural com-
putation and form the algorithmic basis of our content-based information
processing.
Our adaptive solutions are merely variations upon, or extensions of,
previous neural network applications in different fields such as image pro-
cessing, speech recognition, optical character recognition and predictive
coding. For instance, our surface optimization algorithm in Chapter 4
was inspired by the Hidden Markov Model training algorithms used in
speech recognition [Rabiner, 1989]. Our graph comparison algorithm in
Chapter 6 is based on our work on cursive handwriting recognition. The
major contribution of this topic is extending the system design method-
ology that uses neural networks rather than the technology of neural
networks itself. Content-based information processing often is low-level
processing that integrates a high-level feedback mechanism into anal-
ysis. Our solutions for video object extraction and representation are
examples of novel system designs based on adaptive signal and neural
network technologies.
COMPUTER VISION
Although we share many of the computer vision technologies, the goal
of this topic is not to mimic the functionalities human visual system,
but rather to extract and manipulate the visual representations that the
human mind can understand, manipulate and use in communication.
We call these two different goals, a vision-based goal and a media-based
goal. This topic is geared toward a media-based goal.
Toward a vision-based goal, this topic could be considered a minor
field in computer vision. Two main fields of computer vision are already
beyond the scope of this topic: control and feedback to the data acqui-
sition level (active vision) and the use of 3-D reconstruction techniques
(object reconstruction). Our main objectives of this topic are “only”
precursors of computer vision. As mentioned in Section 3., we will deal
with the extraction and representation of video object, not the physical
object itself.
Toward a media-based goal, computer vision plays a supporting role
to our work as implementation technologies and theoretical background.
Computer simulation and study of these biological systems give insight
into the vision process. Motion analysis and image segmentation may
be considered to be a part of computer vision. Much of computer vision
has been devoted to object and scene representation [Ballard and Brown,
1982]. In the next section about video standards, we further distinguish
the computer vision from content-based analysis. While video standards
are a tangential subject for computer vision, they are a key technology in