Digital Signal Processing Reference
In-Depth Information
Fig. 5.5
Visualization of background penalty with occlusion reasoning
Fig. 5.6
Visualization of the smoothness term in “object rectangle”
in Fig.
5.5
b, background penalty with occlusion reasoning introduces a background
penalty factor
α
bp
=
3
.
5 to enforce the background energy of the occluded pixels:
E
p
(
0&
CO
t
(
f
p
)=
α
bp
·
E
p
(
f
p
)(
f
p
=
p
)=
128
)
,
(5.4)
0and
CO
t
(
where
f
p
=
128 if
p
is defined as the occluded background.
E
is the
basic background energy and
E
∗
is the background energy with occlusion penalty.
The motivation of foreground contrast enhancement is to enhance the contrast
across foreground/background boundary and attenuate the background contrast
shown in Fig.
5.6
c by combining the color contrast in L
p
)=
b space in Fig.
5.6
a
and motion residual contrast Fig.
5.6
b. The improved results using modified energy
function is illustrated in Fig.
5.4
d.
×
a
×
Multiview Segmentation
In the above work, we have dealt with the segmentation in a single
key view
of
the
initial frame
. Accurate object segmentation for all views of a frame should be
provided for the further applications.
