Digital Signal Processing Reference
In-Depth Information
TABLE 10.3
Segmentation Results of the Five Contexts Regarding
P
a
,
P
b
, and
P
c
Context-1
Context-2
Context-3
Context-4
Context-5
P
a
0.9365
0.9728
0.9260
0.9186
0.8327
P
b
0.2098
0.3567
0.3716
0.3000
0.1173
P
c
0.6389
0.5189
0.7071
0.7222
0.7237
boundaries that coincide with the true ones, showing
boundary specificity
,
and
P
c
the percentage of true boundaries that can be detected, showing
boundary sensitivity
.Weconduct segmentation experiments on 10 mosaics,
as shown in Figure 10.11. For each context, we perform pixel-level segmen-
tation on the 10 mosaics using the supervised context-based segmentation
algorithm.
7
P
a
,
P
b
, and
P
c
, which are averaged over 10 trials, are shown in
Table 10.3.
A good segmentation requires high
P
a
,
P
b
, and
P
c
. Even though
P
a
is usu-
ally most important, high
P
b
and
P
c
provide more desirable segmentation
results with high accuracy of boundary localization and detection. On the
other hand, boundary localization and detection in textured regions are usu-
ally regarded as difficult issues due to abundant edges and structures around
textured boundaries.
41
,
42
From Table 10.3, it is found that none of the five con-
text models can work well singly in terms of the three criteria. For example,
Context-2 has the best
P
a
but the worst
P
c
. This fact experimentally verifies
that the context models used in References 7 and 33 are good choices in terms
of
P
a
. Context-5 is the strongest in
P
c
but the weakest in
P
a
. Context-3 gives
the highest
P
b
, but
P
a
and
P
c
suffer. These observations are almost completely
consistent in each trial. Intuitively speaking, interscale context models, e.g.,
Context-1 and Context-2, favor
P
a
by encouraging the formation of large, uni-
formly classified regions across scales of the pyramid. The intrascale context
model Context-5 helps
P
c
by being sensitive to boundaries within a scale. As
a hybrid inter- and intrascale context model, Context-3 provides the best
P
b
by appropriately balancing both interscale and intrascale dependencies into
the SMAP Bayesian estimation. Thus, a natural idea is to integrate multiple
context models to achieve high
P
a
,
P
b
, and
P
c
simultaneously.
Generally speaking, given
y
y
(
n
)
|
the collection of mul-
tiscale random fields of an image
Y
,acontext model
V
is used to simplify
the characterization of the joint statistics of
y
with local contextual model-
ing. Thus, given different context models, we can have different statistical
characterizations of
y
. Accordingly, we may have different Bayesian segmen-
tation results. For example, the quadtree pyramid
32
and the interscale context
models
7
,
33
emphasize the homogeneity of the labeling across scales, and the
segmentation results tend to be composed of large, uniformly classified re-
gions. However, those contexts cannot provide high accuracy of boundary
={
n
=
1
,
2
,
...
,L
}
Search WWH ::
Custom Search