EXISTING TECHNIQUES OF VISUAL PROCESSING - Video Object Extraction and Representation: Theory and Applications - page 41

Conceptual Role of Knowledge: a) some points that may be part of an
object boundary or noise, b) if we assume the object boundary goes all points, we are
fitting our boundary to noise. c) if we knew the boundary is smooth, we can reject
noise and find a better boundary estimate.

Figure 2.8.

19961 [Barron et al., 1994] [Aggarwal and Nandhakumar, 1988] [Lai and
Vemuri, 1995]. In our own work, we account for these inaccuracies and
use supplemental motion estimates to avoid this problem.

4. KNOWLEDGE REPRESENTATION
While visual artifacts are important in content-based video process-
ing, a priori object knowledge can be used to refine visual artifacts and
separate content from noise (see Figure 2.8). For the last two sections,
we have focused upon using spatial and spatio-temporal patterns of the
video sequence to infer object features. Our third class of analysis is
based upon the unseen, i.e., not directly derived from I ( x,y,t ) . In this
chapter, we extend our analysis of the video sequence by representing a
priori knowledge of the object and leveraging these representations into
our analysis. Either through assumptions about the qualities of video
objects, generalizations of structure over classes of objects, or a priori
knowledge of the object class, the motion and visual information can be
assembled together in an intelligent manner. This section covers a num-
ber of object representations / models that are available to the designer
and the key issues of how to use these models for content-based analysis.

AVAILABLE REPRESENTATIONS
Representations such as parametric models and neural structures can
explain spatial and temporal patterns within the video sequence as arti-
facts of object structure. What is known can sometimes be more helpful
than what is seen. For instance, when watching analog television, noise
is sometimes seen, but we know to dismiss such visual artifacts on our
television as “bad reception.” We can so easily reject the noise because
television content fits an underlying model in our mind. As shown in Fig-