Digital Signal Processing Reference
In-Depth Information
Chapter 6
A SYSTEM FOR IMAGE/VIDEO OBJECT
QUERY BY SHAPE
In this chapter, we combine our DAG Representation from Chapter 5
and our concept of Voronoi Ordered Space from Chapter 3 to find video
objects in a database that are similar to a given shape. Within the
context of the MPEG-7 standard, tools such as Image
/
Video Query by
Shape (shape query, for short) fulfill an important functionality toward
the creation of a “new” media (see Figure 6.1): a search over terabytes of
Internet video to find relevant video content. After receiving the desired
video content, users can manipulate, blend and arrange this content with
their own content to produce new multimedia documents.
1. PREVIOUS WORK
In this section, we place our work in relation to image query. We
review shape representation methods and link our work to a previous
body of work on Voronoi Skeletons that leverage assumptions about
the object content to model shape deformations [Ogniewicz and Kubler,
1995].
While we consider only the shape characteristics of a single objects in
this chapter, the image query problem searches images that can be com-
posed of many different objects. Image query systems [The and Chin,
1988] [Menon et al., 1996] [Hafner et al., 1995] [Lin et al., 1997] [Smith
and Chang, 1997] [Remias et al., 1996] [Yu and Wolf, 1997] depend
upon the composition of a scene, rather than the properties of a sin-
gle video object. We assume that the video objects are available (e.g.,
from an MPEG-4 stream) while the image query systems must work
on unsegmented images. In this way, we differentiate the image query
problem from shape query problem. While image query blends much
Search WWH ::
Custom Search