Digital Signal Processing Reference
In-Depth Information
the model, which defines image foreground and background histogram distributions
and it is defined as follows.
θ
=
h
(
I
;
α
)
,
α
=
0
,
1
.
(6.1)
Givenanimageframe
I
and model
θ
, the segmentation task is to infer unknown
opacity variables
. The energy function E is defined such that minimum represents
good segmentation and can be formulated as follows.
α
α
=
(
α
,
θ
)
.
argmin
E
(6.2)
α
The model parameters are represented by
)
,
∑
(
α
,
k
)
,
α
=
0
,
1
,
k
=
1
...
K
,
θ
=
π
(
α
,
k
)
,
μ
(
α
,
k
(6.3)
where
is the covariance of 2K Gaussian compo-
nents for foreground and background distributions. Smoothness factor V is defined
as follows.
π
is the weight,
μ
is the mean and
∑
m
,
n
∈
C
[
α
n
2
V
(
α
,
I
)=
γ
=
α
m
]
exp
−
β
||
I
m
−
I
n
||
.
(6.4)
In [
20
], we used scaling function in smoothness factor thereby emphasizing the
importance of depth, which is achieved using weighted L2 norm.
In fact, user initialization can be avoided by using relevant visual attention
models.
6.6
Augmented Reality
Augmented Reality is widely used in many applications. For example, in [
24
], a list
of around thirty interesting augmented reality applications has been discussed. Liu
et al. [
15
] has used model-based video segmentation for interactive games.
6.6.1
Tourism
Depending on the tourist spot, system presents the cultural or heritage story. As
stated in [
27
], system displays the user movement along with narration. Actually,
using segmentation the virtual tourist guide can be associated along with the narra-
tion. This is one of the exciting applications adding real world experience through
mobile devices.
