ROBUST REPRESENTATION OF SHAPE WITH DAGS - Video Object Extraction and Representation: Theory and Applications - page 107

Chapter 5

ROBUST REPRESENTATION OF SHAPE
WITH DAGS

In this chapter, we review our general methodology of robust shape
representation with Directed Acyclic Graphs (DAG S ) [Lin and Kung,
1997a] [Lin and Kung, 1997b] [Lin and Kung, 1998b]. Chapter 2 outlined
the techniques to extract relevant features of the video sequence, Chap-
ter 3 showed the relationship between shape and extracted features, and
Chapter 4 implemented a system to extract video objects themselves. To
search the shapes of extracted video objects, this chapter lays the foun-
dation of the concepts and algorithms for robust shape representation.
Although the examples in this chapter come primarily from our work in
on-line cursive handwriting recognition, the work can be generalized to
drive our video object shape query system in Chapter 6.

1. PREVIOUS WORK
As mentioned in Section 1.9, the search and query functionalities for
video objects proposed by the MPEG-7 standard are closely related to
recognition technologies. In the context of the MPEG-7 standard, the
encoding of video object shape is focused toward a representation that
can be used to identify the class of content of a given video object. Since
the relationship between the shape and the content class is a problem
of not only shape representation, but also content recognition, many
recognition technologies are applicable.
There are many schemes for solving recognition problems as vectors
in a multidimensional feature space [Kung, 1993]. For data with struc-
tured complexity, there are also techniques such as time-dependent neu-
ral nets (TDNN), Hidden Markov Models (HMM) [Bellegarda et al.,
19941 [W. et al., 1995] [M. et al., 1995] [Schenkel et al., 1994] [Garcia-
Salicetti et al., 1995], curve representation techniques [Nishida, 1995]