Digital Signal Processing Reference
In-Depth Information
Figure 6.1.
Image/Video Object Query: a transparent interface layer between users
and Video Databases, translating users requests into database actions and database
references into intelligible results.
representation for a shape is called a
shape descriptor.
The shape query
system implements a
similarity measure
that, given two shapes, returns
a value between 0 and 1, a higher value for two shapes that are more
similar, via comparison of shape descriptors. With this query shape, the
shape query system returns the n shapes that score the
n
-highest sim-
ilarity measures. The user may then refine the search by changing the
query and repeating the search. Although the evaluation of a similarity
measure is affected by assumptions made about the object class (see Fig-
ure 6.2), we can list desirable qualitative properties of shape descriptor
and its similarity measure:
1.
Accuracy
An accurate similarity measure scores the shapes in the same content
class as high and those in different content classes as low. Precision
is secondary: in MPEG-7, we wish to search for classes of objects,
rather than one particular object.
2.
Robustness
Ideally, similarity measures of objects in the same content class should
be invariant to noise, object rotation and object scaling. Since these
shapes exist in a video environment, the similarity measure of objects
in the same content class should also be invariant to different viewing
angles and object movement.
3.
Controllability
User-aided query systems can leverage user feedback to find more
Search WWH ::
Custom Search