EXISTING TECHNIQUES OF VISUAL PROCESSING - Video Object Extraction and Representation: Theory and Applications - page 26

data, but provides insights to the design process. If we accept prepro-
cessing and we are mindful of its biases, preprocessing allows the system
and algorithmic designer to understand and better react to the dynamics
of content-based video processing. This chapter covers techniques from
image segmentation, motion estimation and adaptive signal processing
that are the algorithmic building blocks of our software systems.
This chapter is not meant as a complete study of video and image
processing techniques, but rather a working language of video and image
processing algorithms upon which Chapter 4 and 6 are built. Since we
will be working in terms of these algorithms and their output, these forms
of preprocessing must have reliable qualities: simplicity, robustness, and
graceful degradation. These stringent requirements define a reliable set
of algorithms and techniques that we use in our content-based video
processing.
1. DESIGN IMPLICATIONS OF HUMAN
VISION
An advantage of content-based video processing is that a working
system already exists: our own human visual system. Most problems in
this topic can be easily solved by a well-instructed six-year old child; for
a more personal perspective, you can catalog the objects in a room with
a quick look. Within the human mind, the content-based processing of
object extraction and representation are solved to such a degree that they
are taken for granted. However, if we wish computers to process video,
we still need to translate our own natural abilities into mathematics,
theory, algorithms, and systems.
Our own visual system and pathways guide our research in our content-
based video processing. Since the biological visual system is one of the
most studied parts of human physiology, we can derive much useful in-
formation about how Nature engineered our own vision [Denny, 1994]
[Hubel, 1988] [Marr, 1988]. The physical structure of the retina (see
Figure 2.1a), and cell differentiations and structures (see Figure 2.1b)
demonstrate what cellular specializations are employed for our intake of
visual information. Psychophysical results show the internal processing
of visual input of the human mind by measuring electrical cell impulses
from physical stimuli [Hubel and Wiesel, 1979] [Legge, 19761. Although
the exact mechanisms of the visual pathway are not completely under-
stood, certain dynamics of the human vision can be demonstrated from
psychovisual experiments with reaction time and thresholds of recogni-
tion [Macleod and Rosenfeld, 1974] [Wilson and Bergen, 1979].
Instead of listing and categorizing all the facts, we will summarize the
research work as their relevant implications on our system designs.