Image Processing Reference
In-Depth Information
where from Equation 7.6
2
ϕϕ
() () (,)
x
−
v
=
KxxKvv Kxvi
+
(
, ) (,)
−
2
=
1 2
,
,
…
,
n
and
k
=
1 2
,,
…,
c
i
k
i
i
kk i ik
where
i
is the data point
k
is the cluster centre
Three kernels, namely, the hypertangent kernel, Gaussian kernel and radial
basis kernel are used.
For all these kernels,
K
(
x
i
,
x
i
) = 1 and
K
(
v
k
,
v
k
) = 1.
The objective function in FCM reduces to
n
c
2
∑
m
JUV
(,)
=
u
( (, ))
1
−
Kx v
m
ik
i k
i
=
1
k
=
1
The membership matrix is
(
)
−
1
/(
m
−
1
)
1
−
Kx v
(,)
i k
u
=
ik
∑
c
(
)
−
1
/(
m
−
1
)
1
−
Kx v
(,)
i
j
j
=
1
*
The modified intuitionistic fuzzy membership degree is
uu
ik
=+π
u
ik
is the original membership matrix in FCM where π
ik
= 1 −
u
ik
− (1 −
u
ik
)/
(1 + λ ⋅
u
ik
), and λ
is taken as 1. Sugeno's fuzzy complement is used to compute
the non-membership degree.
The cluster centre after incorporating the intuitionistic property using
hypertangent kernel is written as
;
ik
ik
⎛
⎞
⎛
2
⎞
−−
xv
x
∑
μ
n
*
(,)
k
i
m
⎜
⎜
⎟
⎟
Kx v
1
+
tanh
⎜
⎟
ik
i k
i
2
σ
k
1
⎝
⎠
⎝
⎠
v
=
i
⎛
⎞
⎛
2
⎞
−−
xv
∑
n
*
k
i
m
⎜
⎟
μ
Kx v
(,)
1
+
tanh
⎜
⎟
ik
i k
2
σ
k
=
1
⎝
⎠
⎝
⎠
The cluster centre using Gaussian and radial basis kernel is written as
∑
∑
n
*
(,)
m
μ
Kxvx
ik
i ki
v
=
k
=
1
( 7. 2 4)
i
n
*
(,)
m
μ
Kx v
ik
i k
k
=
1
Search WWH ::
Custom Search