自变量
x
=
[
x
1
,
x
2
]
\mathbf{x}=[x_1,x_2]
x=[x1,x2] 是2维空间中的一个点,函数值
y
=
[
y
1
,
y
2
]
\mathbf{y}=[y_1,y_2]
y=[y1,y2]也是2维空间中的一点,并且都在笛卡尔坐标系下表示。已知存在K个控制点
c
=
{
c
i
1
,
c
i
2
,
i
=
1...
K
}
\mathbf{c}=\{c_{i1},c_{i2},i=1...K\}
c={ci1,ci2,i=1...K} —>
r
=
{
r
i
1
,
r
12
,
i
−
1...
K
}
\mathbf{r}=\{r_{i1},r_{12},i-1...K\}
r={ri1,r12,i−1...K},现给定
N
N
N 个自变量
x
k
\mathbf{x}_k
xk,如何利用薄板样条函数进行插值得到对应的函数值
y
k
\mathbf{y}_k
yk?
设定插值函数为:
Φ
(
x
)
=
[
Φ
1
(
x
)
Φ
2
(
x
)
]
\Phi(\mathbf{x})=\begin{bmatrix} \Phi_1(\mathbf{x}) \\ \Phi_2(\mathbf{x}) \end{bmatrix}
Φ(x)=[Φ1(x)Φ2(x)]
按照薄板样条函数的形式,
Φ
(
x
)
\Phi\mathbf(x)
Φ(x)应为:
Φ
i
(
x
)
=
a
0
+
a
T
x
+
w
T
s
(
x
)
其中
i
=
1
,
2
\Phi_i(\mathbf{x})=a_0+\mathbf{a}^T \mathbf{x}+\mathbf{w}^T \mathbf{s}(\mathbf{x})\quad\quad 其中i=1,2
Φi(x)=a0+aTx+wTs(x)其中i=1,2
展开后,即:
Φ
i
(
x
)
=
a
0
+
[
a
1
,
a
2
]
⋅
[
x
1
x
2
]
+
[
w
1
,
w
2
.
.
.
w
k
]
⋅
[
s
(
x
1
,
x
2
,
c
11
,
c
12
)
s
(
x
1
,
x
2
,
c
21
,
c
22
)
.
.
.
s
(
x
1
,
x
2
,
c
k
1
,
c
k
2
)
]
\Phi_i(\mathbf{x})=a_0+[a_1,a_2]\cdot\begin{bmatrix} x_1\\x_2\end{bmatrix}+[w_1,w_2...w_k]\cdot\begin{bmatrix} s(x_1,x_2,c_{11},c_{12})\\s(x_1,x_2,c_{21},c_{22})\\...\\s(x_1,x_2,c_{k1},c_{k2})\end{bmatrix}
Φi(x)=a0+[a1,a2]⋅[x1x2]+[w1,w2...wk]⋅
s(x1,x2,c11,c12)s(x1,x2,c21,c22)...s(x1,x2,ck1,ck2)
由于
σ
(
r
)
=
r
2
log
(
r
)
\sigma(r)=r^2\log(r)
σ(r)=r2log(r), 所以:
s
(
x
1
,
x
2
,
c
i
1
,
c
i
2
)
=
[
(
x
1
−
c
i
1
)
2
+
(
x
2
−
c
i
2
)
2
]
×
log
(
[
(
x
1
−
c
i
1
)
2
+
(
x
2
−
c
i
2
)
2
]
)
s(x_1,x_2,c_{i1},c_{i2})=[(x_1-c_{i1})^2+(x_2-c_{i2})^2]\times\log(\sqrt{[(x_1-c_{i1})^2+(x_2-c_{i2})^2])}
s(x1,x2,ci1,ci2)=[(x1−ci1)2+(x2−ci2)2]×log([(x1−ci1)2+(x2−ci2)2])
根据相关推导1,公式中的参数需满足以下约束条件:
∑
i
=
1
K
w
i
=
0
\sum_{i=1}^{\mathbf{K}}w_i=0
i=1∑Kwi=0
∑
i
=
1
K
c
i
1
w
i
=
0
\sum_{i=1}^{\mathbf{K}}c_{i1}w_i=0
i=1∑Kci1wi=0
∑
i
=
1
K
c
i
2
w
i
=
0
\sum_{i=1}^{\mathbf{K}}c_{i2}w_i=0
i=1∑Kci2wi=0
对于
Φ
1
(
x
)
\Phi_1(\mathbf{x})
Φ1(x)的参数,最终形成如下求解方程:
[
1
c
11
c
12
s
(
c
11
,
c
12
,
c
11
,
c
12
)
s
(
c
11
,
c
12
,
c
21
,
c
22
)
.
.
.
s
(
c
11
,
c
12
,
c
k
1
,
c
k
2
)
1
c
21
c
22
s
(
c
21
,
c
22
,
c
11
,
c
12
)
s
(
c
21
,
c
22
,
c
21
,
c
22
)
.
.
.
s
(
c
21
,
c
22
,
c
k
1
,
c
k
2
)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
c
k
1
c
k
2
s
(
c
k
1
,
c
k
2
,
c
11
,
c
12
)
s
(
c
k
1
,
c
k
2
,
c
21
,
c
22
)
.
.
.
s
(
c
k
1
,
c
k
2
,
c
k
1
,
c
k
2
)
0
0
0
1
1
.
.
.
1
0
0
0
c
11
c
21
.
.
.
c
k
1
0
0
0
c
12
c
22
.
.
.
c
k
2
]
[
a
0
a
1
a
2
w
1
w
2
.
.
.
w
k
]
=
[
r
12
r
21
.
.
.
r
k
1
0
0
0
]
\begin{bmatrix} 1&c_{11}&c_{12}&s(c_{11},c_{12},c_{11},c_{12})&s(c_{11},c_{12},c_{21},c_{22})&...&s(c_{11},c_{12},c_{k1},c_{k2})\\1&c_{21}&c_{22}&s(c_{21},c_{22},c_{11},c_{12})&s(c_{21},c_{22},c_{21},c_{22})&...&s(c_{21},c_{22},c_{k1},c_{k2})\\\quad&\quad&...&...&...&...&...\\1&c_{k1}&c_{k2}&s(c_{k1},c_{k2},c_{11},c_{12})&s(c_{k1},c_{k2},c_{21},c_{22})&...&s(c_{k1},c_{k2},c_{k1},c_{k2})\\0&0&0&1&1&...&1\\0&0&0&c_{11}&c_{21}&...&c_{k1}\\0&0&0&c_{12}&c_{22}&...&c_{k2}\end{bmatrix}\begin{bmatrix}a_0\\a_1\\a_2\\w_1\\w_2\\...\\w_k\end{bmatrix}=\begin{bmatrix}r_{12}\\r_{21}\\...\\r_{k1}\\0\\0\\0\end{bmatrix}
111000c11c21ck1000c12c22...ck2000s(c11,c12,c11,c12)s(c21,c22,c11,c12)...s(ck1,ck2,c11,c12)1c11c12s(c11,c12,c21,c22)s(c21,c22,c21,c22)...s(ck1,ck2,c21,c22)1c21c22.....................s(c11,c12,ck1,ck2)s(c21,c22,ck1,ck2)...s(ck1,ck2,ck1,ck2)1ck1ck2
a0a1a2w1w2...wk
=
r12r21...rk1000
其中:
s
(
c
i
1
,
c
i
2
,
c
i
1
,
c
i
2
)
=
0
s(c_{i1},c_{i2},c_{i1},c_{i2})=0
s(ci1,ci2,ci1,ci2)=0
同理可以计算得到
Φ
2
(
x
)
\Phi_2(\mathbf{x})
Φ2(x)
https://www.jianshu.com/p/2cc189dfbcc5#fn3 ↩︎