Digital Signal Processing Reference
In-Depth Information
show the drawbacks and trade-offs that balance our functionality. Our
results confirm the system functionality and show the system is com-
petitive with established methods of shape descriptors such as curvature
space, Zernike moments and Fourier descriptors. Although the interpre-
tation of such results is always subjective with content-based technolo-
gies, we wish to claim that our shape query systems has good enough
results to
validate its existence,
while demonstrating its novel function-
ality.
DATABASE #1: SELECTIONS FROM ANIMAL
DATABASE
The first test database is a small database of selected 12 animals
to demonstrate functionalities and the future possibilities of VOS-DOT
shape query system. This small database is biased toward our system
due to the fact that most animals follow our assumptions of implicit
support structure.
Our novel functionality is validated through these results upon the
small database. From results in Figure 6.14, the first query (
) matches
with the shape A, even though the query is rotated in 3-D space. For-
tunately, in this case, the 3-D rotation does not affect the projection of
physical skeleton and the similarity measure matches the two descrip-
tors well.
α
The second query (
β
) matches well with H and clearly rejects
all other
animals due to the
six-legged
structure
of the beetle.
For an
unknown animal query
, an impala), our similarity measure finds the
closest match to the horse shape through the removal of the horns.
(
γ
Fi-
) has very different shape than (D),
the skeletons remain constant and the similarity measure matches them
well. From this example, the skeleton separates motion and structural
information, allowing the user to annotate motion on its shape descrip-
tor. Second and third matches can be explained through deformations of
structure, intuitive (C,
nally, although the jumping horse (
δ
δ
) and non-intuitive (D,
α
), or penalized removal
of extremities
). In this experiment, the situations were obviously
controlled in our favor. Our next database has a more objective criterion
for correctness and has a greater range of shapes to show how the VOS
representation aliases and in which circumstances this shape descriptor
is advantageous and disadvantageous.
(F,
β
DATABASE #2: MPEG-7 CORE EXPERIMENT
ON SHAPE
The second large database of approximately 1400 shapes shows the
effects of aliasing and an intuition about the behavior of our similarity
Search WWH ::
Custom Search