EXISTING TECHNIQUES OF VISUAL PROCESSING - Video Object Extraction and Representation: Theory and Applications - page 29

Fourier coefficients since we achieve good compression rates and good
psychovisual quality with frequency-based techniques. The choice of
domain greatly affects how we design our systems. Visual testing and
biological structure leans heavily toward a representation based on
spatial coordinates. Our concept of video space coordinates ( x, y, t )
match well with how the retina receives the visual information.

4. Vision Supports the Differential as a Basic Operation

The human visual system has an amazing visual operating range that
spans over 10-11 logarithmic scales of light intensity. The key to the
broad operating range is that the retinal structure implements a local
differential operation by placing the two different types of cells within
an excitory-inhibitory structure (see Figure 2.1b). For instance, a
green-red difference is implemented by placing a green cone within
an excitory region surrounded by red cones in an inhibitory region.
The spatial difference operation is clearly a basic visual operation of
the eye.

5. Vision is a Temporally Linked Process

The retina has its own transient state. Although we can consider the
video sequence as series of images, an isolated image does not have
an equivalent effect of an image within the context of the sequence.
This effect of previously seen images on the retina can last beyond
100 milliseconds. Time is not only a construct of the mind, but also
a physical attribute of visual acquisition devices in our retina. These
effects show the importance and scale of temporal correlation that
exists between sequential frames.

In this section, we have outlined some surprising features of our own
visual system. We considered only low-level processing because 1) we
wished to avoid misinterpretation of the biological results and 2) the
bulk of the research work and understanding of vision is concentrated
at the cellular level. Although few insights into the high-level processes
of human vision are given, we can understand some of biological visual
preprocessing.

2. IMAGE SEGMENTATION

Image segmentation is the precursor to content-based information pro-
cessing. Image segmentation is the extraction of VOPs from a single
frame of video. By extending these image segmentation techniques to
handle multiple frames, we can begin to design our content-based video
processing systems. Furthermore, image segmentation has many com-
mon issues as our content-based video processing: duality in problem