數學上,任一的
M
×
N
{\displaystyle M\times N}
離散線性轉換皆可表示成矩陣 (Matrix) 的型式:
Y
=
A
X
{\displaystyle {\boldsymbol {Y=AX}}}
[
y
[
0
]
y
[
1
]
y
[
2
]
⋮
y
[
M
−
1
]
]
=
[
a
0
[
0
]
a
0
[
1
]
a
0
[
2
]
⋯
a
0
[
N
−
1
]
a
1
[
0
]
a
1
[
1
]
a
1
[
2
]
⋯
a
1
[
N
−
1
]
a
2
[
0
]
a
2
[
1
]
a
2
[
2
]
⋯
a
2
[
N
−
1
]
⋮
⋮
⋮
⋱
⋮
a
M
−
1
[
0
]
a
M
−
1
[
1
]
a
M
−
1
[
2
]
⋯
a
M
−
1
[
N
−
1
]
]
[
x
[
0
]
x
[
1
]
x
[
2
]
⋮
x
[
N
−
1
]
]
.
{\displaystyle {\begin{bmatrix}y\left[{0}\right]\\y\left[{1}\right]\\y\left[{2}\right]\\\vdots \\y\left[{M-1}\right]\end{bmatrix}}={\begin{bmatrix}{a_{0}\left[0\right]}&{a_{0}\left[1\right]}&{a_{0}\left[2\right]}&\cdots &{a_{0}\left[N-1\right]}\\{a_{1}\left[0\right]}&{a_{1}\left[1\right]}&{a_{1}\left[2\right]}&\cdots &{a_{1}\left[N-1\right]}\\{a_{2}\left[0\right]}&{a_{2}\left[1\right]}&{a_{2}\left[2\right]}&\cdots &{a_{2}\left[N-1\right]}\\\vdots &\vdots &\vdots &\ddots &\vdots \\{a_{M-1}\left[0\right]}&{a_{M-1}\left[1\right]}&{a_{M-1}\left[2\right]}&\cdots &{a_{M-1}\left[N-1\right]}\end{bmatrix}}{\begin{bmatrix}x\left[{0}\right]\\x\left[{1}\right]\\x\left[{2}\right]\\\vdots \\x\left[{N-1}\right]\end{bmatrix}}.}
C
A
S
E
1
{\displaystyle \mathbf {{CASE}\,{1}} }
再進一步假設,若矩陣
A
{\displaystyle {\boldsymbol {A}}}
by正交基底 (Orthogonal basis) 列向量 (Row vector) 所組成:
[
ϕ
0
∗
ϕ
1
∗
ϕ
2
∗
⋮
ϕ
M
−
1
∗
]
=
[
ϕ
0
∗
[
0
]
ϕ
0
∗
[
1
]
ϕ
0
∗
[
2
]
⋯
ϕ
0
∗
[
N
−
1
]
ϕ
1
∗
[
0
]
ϕ
1
∗
[
1
]
ϕ
1
∗
[
2
]
⋯
ϕ
1
∗
[
N
−
1
]
ϕ
2
∗
[
0
]
ϕ
2
∗
[
1
]
ϕ
2
∗
[
2
]
⋯
ϕ
2
∗
[
N
−
1
]
⋮
⋮
⋮
⋮
⋮
ϕ
M
−
1
∗
[
0
]
ϕ
M
−
1
∗
[
1
]
ϕ
M
−
1
∗
[
2
]
⋯
ϕ
M
−
1
∗
[
N
−
1
]
]
=
A
.
{\displaystyle {\begin{bmatrix}{{\boldsymbol {\phi }}_{0}^{*}}\\{{\boldsymbol {\phi }}_{1}^{*}}\\{{\boldsymbol {\phi }}_{2}^{*}}\\\vdots \\{{\boldsymbol {\phi }}_{M-1}^{*}}\end{bmatrix}}={\begin{bmatrix}{\phi _{0}^{*}\left[0\right]}&{\phi _{0}^{*}\left[1\right]}&{\phi _{0}^{*}\left[2\right]}&\cdots &{\phi _{0}^{*}\left[N-1\right]}\\{\phi _{1}^{*}\left[0\right]}&{\phi _{1}^{*}\left[1\right]}&{\phi _{1}^{*}\left[2\right]}&\cdots &{\phi _{1}^{*}\left[N-1\right]}\\{\phi _{2}^{*}\left[0\right]}&{\phi _{2}^{*}\left[1\right]}&{\phi _{2}^{*}\left[2\right]}&\cdots &{\phi _{2}^{*}\left[N-1\right]}\\\vdots &\vdots &\vdots &\vdots &\vdots \\{\phi _{M-1}^{*}\left[0\right]}&{\phi _{M-1}^{*}\left[1\right]}&{\phi _{M-1}^{*}\left[2\right]}&\cdots &{\phi _{M-1}^{*}\left[N-1\right]}\end{bmatrix}}={\boldsymbol {A}}.}
也可表示成級數和形式(Summation form):
y
[
m
]
=
⟨
ϕ
m
∗
,
x
[
n
]
⟩
=
∑
n
=
0
N
−
1
ϕ
m
∗
[
n
]
x
[
n
]
,
for
m
=
0
,
1
,
⋯
,
M
−
1.
{\displaystyle y\left[m\right]=\left\langle {{\boldsymbol {\phi }}_{m}^{*}},x\left[n\right]\right\rangle =\sum _{n=0}^{N-1}{\phi _{m}^{*}\left[n\right]}x\left[n\right],\quad {\mbox{for }}m=0,\,1,\,\cdots ,\,M-1.}
其中
⟨
⋅
,
⋅
⟩
{\displaystyle \left\langle \cdot \,,\cdot \right\rangle }
代表內積 運算(Inner product)。
C
A
S
E
2
{\displaystyle \mathbf {{CASE}\,{2}} }
同理也可假設,若矩陣
A
{\displaystyle {\boldsymbol {A}}}
by正交基底行向量 (Column vector) 所組成:
[
β
0
∗
β
1
∗
β
2
∗
⋯
β
N
−
1
∗
]
=
[
β
0
∗
[
0
]
β
1
∗
[
0
]
β
2
∗
[
0
]
⋯
β
N
−
1
∗
[
0
]
β
0
∗
[
1
]
β
1
∗
[
1
]
β
2
∗
[
1
]
⋯
β
N
−
1
∗
[
1
]
β
0
∗
[
2
]
β
1
∗
[
2
]
β
2
∗
[
2
]
⋯
β
N
−
1
∗
[
2
]
⋮
⋮
⋮
⋮
⋮
β
0
∗
[
M
−
1
]
β
1
∗
[
M
−
1
]
β
2
∗
[
M
−
1
]
⋯
β
N
−
1
∗
[
M
−
1
]
]
=
A
.
{\displaystyle {\begin{bmatrix}{{\boldsymbol {\beta }}_{0}^{*}}&{{\boldsymbol {\beta }}_{1}^{*}}&{{\boldsymbol {\beta }}_{2}^{*}}&\cdots &{{\boldsymbol {\beta }}_{N-1}^{*}}\end{bmatrix}}={\begin{bmatrix}{\beta _{0}^{*}\left[0\right]}&{\beta _{1}^{*}\left[0\right]}&{\beta _{2}^{*}\left[0\right]}&\cdots &{\beta _{N-1}^{*}\left[0\right]}\\{\beta _{0}^{*}\left[1\right]}&{\beta _{1}^{*}\left[1\right]}&{\beta _{2}^{*}\left[1\right]}&\cdots &{\beta _{N-1}^{*}\left[1\right]}\\{\beta _{0}^{*}\left[2\right]}&{\beta _{1}^{*}\left[2\right]}&{\beta _{2}^{*}\left[2\right]}&\cdots &{\beta _{N-1}^{*}\left[2\right]}\\\vdots &\vdots &\vdots &\vdots &\vdots \\{\beta _{0}^{*}\left[M-1\right]}&{\beta _{1}^{*}\left[M-1\right]}&{\beta _{2}^{*}\left[M-1\right]}&\cdots &{\beta _{N-1}^{*}\left[M-1\right]}\end{bmatrix}}={\boldsymbol {A}}.}
也可表示成級數和:
y
[
m
]
=
∑
n
=
0
N
−
1
x
[
n
]
β
n
∗
[
m
]
,
for
m
=
0
,
1
,
⋯
,
M
−
1.
{\displaystyle y\left[m\right]=\sum _{n=0}^{N-1}x\left[n\right]{\beta _{n}^{*}\left[m\right]},\quad {\mbox{for }}m=0,\,1,\,\cdots ,\,M-1.}
假若
M
=
N
{\displaystyle M=N}
時,則
d
e
t
(
A
)
≠
0
{\displaystyle det(A)\neq 0}
可得
{
Y
=
A
X
(
Forward transform
)
X
=
A
−
1
X
(
Inverse transform
)
.
{\displaystyle {\begin{cases}Y=AX&({\mbox{Forward transform}})\\X=A^{-1}X&({\mbox{Inverse transform}})\end{cases}}.}
大致上,可簡單化將矩陣 分類由(a)列向量 (Row Vector)或(b)行向量 (Column Vector)所組成。
正交矩陣by列向量:
A
=
[
ϕ
0
∗
ϕ
1
∗
ϕ
2
∗
⋮
ϕ
M
−
1
∗
]
{\displaystyle {\boldsymbol {A}}={\begin{bmatrix}{{\boldsymbol {\phi }}_{0}^{*}}\\{{\boldsymbol {\phi }}_{1}^{*}}\\{{\boldsymbol {\phi }}_{2}^{*}}\\\vdots \\{{\boldsymbol {\phi }}_{M-1}^{*}}\end{bmatrix}}}
若
{
⟨
ϕ
l
∗
,
ϕ
k
∗
⟩
=
∑
n
=
0
N
−
1
ϕ
l
∗
[
n
]
(
ϕ
k
∗
[
n
]
)
T
=
0
,
when
l
≠
k
⟨
ϕ
l
∗
,
ϕ
l
∗
⟩
=
∑
n
=
0
N
−
1
ϕ
l
∗
[
n
]
(
ϕ
l
∗
[
n
]
)
T
=
C
l
=
C
o
n
s
t
a
n
t
,
{\displaystyle {\begin{cases}\;\left\langle {\boldsymbol {\phi }}_{l}^{*},{\boldsymbol {\phi }}_{k}^{*}\right\rangle =\sum _{n=0}^{N-1}{\phi _{l}^{*}\left[n\right]}\left({\phi _{k}^{*}\left[n\right]}\right)^{T}=0,\quad {\mbox{when }}l\neq k\\\;\left\langle {\boldsymbol {\phi }}_{l}^{*},{\boldsymbol {\phi }}_{l}^{*}\right\rangle =\sum _{n=0}^{N-1}{\phi _{l}^{*}\left[n\right]}\left({\phi _{l}^{*}\left[n\right]}\right)^{T}=C_{l}=Constant\end{cases}},}
其中,
ϕ
0
∗
,
ϕ
1
∗
,
ϕ
2
∗
,
…
,
{\displaystyle {\;{\boldsymbol {\phi }}_{0}^{*}},\;{{\boldsymbol {\phi }}_{1}^{*}},\;{{\boldsymbol {\phi }}_{2}^{*}},\;\ldots ,\;}
和
ϕ
M
−
1
∗
{\displaystyle {\;{\boldsymbol {\phi }}_{M-1}^{*}}}
為一組列向量 的正交集合且稱
A
{\displaystyle {\boldsymbol {A}}\ }
為一種離散正交轉換。
再者,若滿足
C
l
=
1
,
for any
l
{\displaystyle C_{l}=1,\;{\mbox{for any }}l}
。則
ϕ
0
∗
,
ϕ
1
∗
,
ϕ
2
∗
,
…
,
{\displaystyle {\;{\boldsymbol {\phi }}_{0}^{*}},\;{{\boldsymbol {\phi }}_{1}^{*}},\;{{\boldsymbol {\phi }}_{2}^{*}},\;\ldots ,\;}
和
ϕ
M
−
1
∗
{\displaystyle {\;{\boldsymbol {\phi }}_{M-1}^{*}}}
將組成一組列向量 的正規化正交集合且稱
A
{\displaystyle {\boldsymbol {A}}\ }
為一種離散正規化正交轉換。
此時,我們可利用
ϕ
0
∗
,
ϕ
1
∗
,
ϕ
2
∗
,
…
,
{\displaystyle {\;{\boldsymbol {\phi }}_{0}^{*}},\;{{\boldsymbol {\phi }}_{1}^{*}},\;{{\boldsymbol {\phi }}_{2}^{*}},\;\ldots ,\;}
和
ϕ
M
−
1
∗
{\displaystyle {\;{\boldsymbol {\phi }}_{M-1}^{*}}}
來求得
A
{\displaystyle {\boldsymbol {A}}}
的反矩陣 :
A
−
1
=
[
C
0
−
1
ϕ
0
∗
[
0
]
C
1
−
1
ϕ
1
∗
[
0
]
C
2
−
1
ϕ
2
∗
[
0
]
⋯
C
N
−
1
−
1
ϕ
N
−
1
∗
[
0
]
C
0
−
1
ϕ
0
∗
[
1
]
C
1
−
1
ϕ
1
∗
[
1
]
C
2
−
1
ϕ
2
∗
[
1
]
⋯
C
N
−
1
−
1
ϕ
N
−
1
∗
[
1
]
C
0
−
1
ϕ
0
∗
[
2
]
C
1
−
1
ϕ
1
∗
[
2
]
C
2
−
1
ϕ
2
∗
[
2
]
⋯
C
N
−
1
−
1
ϕ
N
−
1
∗
[
2
]
⋮
⋮
⋮
⋱
⋮
C
0
−
1
ϕ
0
∗
[
M
−
1
]
C
1
−
1
ϕ
1
∗
[
M
−
1
]
C
2
−
1
ϕ
2
∗
[
M
−
1
]
⋯
C
N
−
1
−
1
ϕ
N
−
1
∗
[
M
−
1
]
]
,
{\displaystyle {\boldsymbol {A}}^{-1}={\begin{bmatrix}{C_{0}^{-1}\phi _{0}^{*}\left[0\right]}&{C_{1}^{-1}\phi _{1}^{*}\left[0\right]}&{C_{2}^{-1}\phi _{2}^{*}\left[0\right]}&\cdots &{C_{N-1}^{-1}\phi _{N-1}^{*}\left[0\right]}\\{C_{0}^{-1}\phi _{0}^{*}\left[1\right]}&{C_{1}^{-1}\phi _{1}^{*}\left[1\right]}&{C_{2}^{-1}\phi _{2}^{*}\left[1\right]}&\cdots &{C_{N-1}^{-1}\phi _{N-1}^{*}\left[1\right]}\\{C_{0}^{-1}\phi _{0}^{*}\left[2\right]}&{C_{1}^{-1}\phi _{1}^{*}\left[2\right]}&{C_{2}^{-1}\phi _{2}^{*}\left[2\right]}&\cdots &{C_{N-1}^{-1}\phi _{N-1}^{*}\left[2\right]}\\\vdots &\vdots &\vdots &\ddots &\vdots \\{C_{0}^{-1}\phi _{0}^{*}\left[M-1\right]}&{C_{1}^{-1}\phi _{1}^{*}\left[M-1\right]}&{C_{2}^{-1}\phi _{2}^{*}\left[M-1\right]}&\cdots &{C_{N-1}^{-1}\phi _{N-1}^{*}\left[M-1\right]}\end{bmatrix}},}
其中
C
m
=
⟨
ϕ
m
∗
,
ϕ
m
∗
⟩
{\displaystyle C_{m}=\left\langle {\boldsymbol {\phi }}_{m}^{*},{\boldsymbol {\phi }}_{m}^{*}\right\rangle }
。再者,也可表示成級數和(Summation)形式:
x
[
n
]
=
∑
m
=
0
N
−
1
C
m
−
1
ϕ
m
∗
[
n
]
y
[
m
]
,
for
n
=
0
,
1
,
⋯
,
N
−
1.
{\displaystyle x\left[{n}\right]=\sum _{m=0}^{N-1}C_{m}^{-1}{\phi _{m}^{*}\left[n\right]}y\left[{m}\right],\quad {\mbox{for }}n=0,\,1,\,\cdots ,\,N-1.}
若
C
m
=
1
{\displaystyle C_{m}=1}
,即簡化成,
x
[
n
]
=
∑
m
=
0
N
−
1
ϕ
m
∗
[
n
]
y
[
m
]
,
for
n
=
0
,
1
,
⋯
,
N
−
1.
{\displaystyle x\left[{n}\right]=\sum _{m=0}^{N-1}{\phi _{m}^{*}\left[n\right]}y\left[{m}\right],\quad {\mbox{for }}n=0,\,1,\,\cdots ,\,N-1.}
正交矩陣by行向量:
A
=
[
β
0
∗
β
1
∗
β
2
∗
⋯
β
N
−
1
∗
]
{\displaystyle {\boldsymbol {A}}={\begin{bmatrix}{{\boldsymbol {\beta }}_{0}^{*}}&{{\boldsymbol {\beta }}_{1}^{*}}&{{\boldsymbol {\beta }}_{2}^{*}}&\cdots &{{\boldsymbol {\beta }}_{N-1}^{*}}\end{bmatrix}}}
若
{
⟨
β
l
∗
,
β
k
∗
⟩
=
∑
m
=
0
M
−
1
(
β
l
∗
[
m
]
)
T
β
k
∗
[
m
]
=
0
,
when
l
≠
k
⟨
β
l
∗
,
β
l
∗
⟩
=
∑
m
=
0
M
−
1
(
β
l
∗
[
m
]
)
T
β
l
∗
[
m
]
=
C
l
=
C
o
n
s
t
a
n
t
,
{\displaystyle {\begin{cases}\;\left\langle {\boldsymbol {\beta }}_{l}^{*},{\boldsymbol {\beta }}_{k}^{*}\right\rangle =\sum _{m=0}^{M-1}\left({\beta _{l}^{*}\left[m\right]}\right)^{T}{\beta _{k}^{*}\left[m\right]}=0,\quad {\mbox{when }}l\neq k\\\;\left\langle {\boldsymbol {\beta }}_{l}^{*},{\boldsymbol {\beta }}_{l}^{*}\right\rangle =\sum _{m=0}^{M-1}\left({\beta _{l}^{*}\left[m\right]}\right)^{T}{\beta _{l}^{*}\left[m\right]}=C_{l}=Constant\end{cases}},}
其中,
β
0
∗
,
β
1
∗
,
β
2
∗
,
…
,
{\displaystyle {\;{\boldsymbol {\beta }}_{0}^{*}},\;{{\boldsymbol {\beta }}_{1}^{*}},\;{{\boldsymbol {\beta }}_{2}^{*}},\;\ldots ,\;}
和
β
N
−
1
∗
{\displaystyle {\;{\boldsymbol {\beta }}_{N-1}^{*}}}
為一組行向量 的正交集合且也稱
A
{\displaystyle {\boldsymbol {A}}\ }
是一種離散正交轉換。
再者,若滿足
C
l
=
1
,
for any
l
{\displaystyle C_{l}=1,\;{\mbox{for any }}l}
。則
β
0
∗
,
β
1
∗
,
β
2
∗
,
…
,
{\displaystyle {\;{\boldsymbol {\beta }}_{0}^{*}},\;{{\boldsymbol {\beta }}_{1}^{*}},\;{{\boldsymbol {\beta }}_{2}^{*}},\;\ldots ,\;}
和
β
N
−
1
∗
{\displaystyle {\;{\boldsymbol {\beta }}_{N-1}^{*}}}
將組成一組行向量 的正規化正交集合且稱
A
{\displaystyle {\boldsymbol {A}}\ }
為一種離散正規化正交轉換。
此時,我們再利用
β
0
∗
,
β
1
∗
,
β
2
∗
,
…
,
{\displaystyle {\;{\boldsymbol {\beta }}_{0}^{*}},\;{{\boldsymbol {\beta }}_{1}^{*}},\;{{\boldsymbol {\beta }}_{2}^{*}},\;\ldots ,\;}
和
β
N
−
1
∗
{\displaystyle {\;{\boldsymbol {\beta }}_{N-1}^{*}}}
來組成
A
{\displaystyle {\boldsymbol {A}}}
的反矩陣 :
A
−
1
=
[
C
0
−
1
β
0
∗
[
0
]
C
0
−
1
β
0
∗
[
1
]
C
0
−
1
β
0
∗
[
2
]
⋯
C
0
−
1
β
0
∗
[
N
−
1
]
C
1
−
1
β
1
∗
[
0
]
C
1
−
1
β
1
∗
[
1
]
C
1
−
1
β
1
∗
[
2
]
⋯
C
1
−
1
β
1
∗
[
N
−
1
]
C
2
−
1
β
2
∗
[
0
]
C
2
−
1
β
2
∗
[
1
]
C
2
−
1
β
2
∗
[
2
]
⋯
C
2
−
1
β
2
∗
[
N
−
1
]
⋮
⋮
⋮
⋱
⋮
C
M
−
1
−
1
β
M
−
1
∗
[
0
]
C
M
−
1
−
1
β
M
−
1
∗
[
1
]
C
M
−
1
−
1
β
M
−
1
∗
[
2
]
⋯
C
M
−
1
−
1
β
M
−
1
∗
[
N
−
1
]
]
,
{\displaystyle {\boldsymbol {A}}^{-1}={\begin{bmatrix}{C_{0}^{-1}\beta _{0}^{*}\left[0\right]}&{C_{0}^{-1}\beta _{0}^{*}\left[1\right]}&{C_{0}^{-1}\beta _{0}^{*}\left[2\right]}&\cdots &{C_{0}^{-1}\beta _{0}^{*}\left[N-1\right]}\\{C_{1}^{-1}\beta _{1}^{*}\left[0\right]}&{C_{1}^{-1}\beta _{1}^{*}\left[1\right]}&{C_{1}^{-1}\beta _{1}^{*}\left[2\right]}&\cdots &{C_{1}^{-1}\beta _{1}^{*}\left[N-1\right]}\\{C_{2}^{-1}\beta _{2}^{*}\left[0\right]}&{C_{2}^{-1}\beta _{2}^{*}\left[1\right]}&{C_{2}^{-1}\beta _{2}^{*}\left[2\right]}&\cdots &{C_{2}^{-1}\beta _{2}^{*}\left[N-1\right]}\\\vdots &\vdots &\vdots &\ddots &\vdots \\{C_{M-1}^{-1}\beta _{M-1}^{*}\left[0\right]}&{C_{M-1}^{-1}\beta _{M-1}^{*}\left[1\right]}&{C_{M-1}^{-1}\beta _{M-1}^{*}\left[2\right]}&\cdots &{C_{M-1}^{-1}\beta _{M-1}^{*}\left[N-1\right]}\end{bmatrix}},}
其中
C
n
=
⟨
β
n
∗
,
β
n
∗
⟩
{\displaystyle C_{n}=\left\langle {\boldsymbol {\beta }}_{n}^{*},{\boldsymbol {\beta }}_{n}^{*}\right\rangle }
。再者,也可表示成級數和:
x
[
n
]
=
C
n
−
1
∑
m
=
0
N
−
1
β
n
∗
[
m
]
y
[
m
]
,
for
n
=
0
,
1
,
⋯
,
N
−
1.
{\displaystyle x\left[{n}\right]=C_{n}^{-1}\sum _{m=0}^{N-1}{\beta _{n}^{*}\left[m\right]}y\left[{m}\right],\quad {\mbox{for }}n=0,\,1,\,\cdots ,\,N-1.}
若
C
n
=
1
{\displaystyle C_{n}=1}
,即簡化成,
x
[
n
]
=
∑
m
=
0
N
−
1
β
n
∗
[
m
]
y
[
m
]
,
for
n
=
0
,
1
,
⋯
,
N
−
1.
{\displaystyle x\left[{n}\right]=\sum _{m=0}^{N-1}{\beta _{n}^{*}\left[m\right]}y\left[{m}\right],\quad {\mbox{for }}n=0,\,1,\,\cdots ,\,N-1.}
通常對於影像重建 或壓縮 上大都採用局部重建(Parital reconstruction)的機制,即:
(a) 正交 情況下
x
k
[
n
]
=
∑
m
=
0
K
−
1
C
m
−
1
y
[
m
]
ϕ
m
∗
[
n
]
,
for
K
<
N
.
{\displaystyle x_{k}\left[{n}\right]=\sum _{m=0}^{K-1}C_{m}^{-1}y\left[{m}\right]{\phi _{m}^{*}\left[n\right]},\quad {\mbox{for }}K<N.}
因此,對於局部重建所產生的平方誤差 (Sqaure error):
‖
x
[
n
]
−
x
k
[
n
]
‖
2
=
∑
n
=
0
N
−
1
‖
∑
m
=
k
N
−
1
C
m
−
1
y
[
m
]
ϕ
m
[
n
]
‖
2
=
∑
n
=
0
N
−
1
∑
m
=
k
N
−
1
C
m
−
1
y
[
m
]
ϕ
m
[
n
]
∑
h
=
k
N
−
1
C
h
−
1
y
∗
[
h
]
ϕ
h
∗
[
n
]
=
∑
m
=
k
N
−
1
C
m
−
1
|
y
[
m
]
|
2
.
{\displaystyle \left\|x\left[n\right]-x_{k}\left[n\right]\right\|^{2}=\sum _{n=0}^{N-1}\left\|\sum _{m=k}^{N-1}C_{m}^{-1}y\left[{m}\right]\phi _{m}\left[n\right]\right\|^{2}=\sum _{n=0}^{N-1}\sum _{m=k}^{N-1}C_{m}^{-1}y\left[{m}\right]\phi _{m}\left[n\right]\sum _{h=k}^{N-1}C_{h}^{-1}y^{*}\left[{h}\right]\phi _{h}^{*}\left[n\right]=\sum _{m=k}^{N-1}C_{m}^{-1}\left|\ y\left[{m}\right]\right|^{2}.}
從此結果可發現
C
m
−
1
|
y
[
m
]
|
2
{\displaystyle C_{m}^{-1}\left|\ y\left[{m}\right]\right|^{2}}
必定為正的,因此可藉由增加正交基底數來改善影像重建後的平方誤差值。
(b)非正交情況下
x
k
[
n
]
=
∑
m
=
0
K
−
1
B
[
n
,
m
]
y
[
m
]
,
for
K
<
N
.
{\displaystyle x_{k}\left[{n}\right]=\sum _{m=0}^{K-1}B\left[{n,m}\right]y\left[m\right],\quad {\mbox{for }}K<N.}
因此,對於局部重建所產生的平方誤差 :
‖
x
[
n
]
−
x
k
[
n
]
‖
2
=
∑
n
=
0
N
−
1
‖
∑
m
=
k
N
−
1
B
[
n
,
m
]
y
[
m
]
‖
2
=
∑
m
=
k
N
−
1
∑
h
=
k
N
−
1
y
[
m
]
y
∗
[
h
]
∑
n
=
0
N
−
1
B
[
n
,
m
]
B
∗
[
n
,
h
]
.
{\displaystyle \left\|x\left[n\right]-x_{k}\left[n\right]\right\|^{2}=\sum _{n=0}^{N-1}\left\|\sum _{m=k}^{N-1}B\left[{n,m}\right]y\left[m\right]\right\|^{2}=\sum _{m=k}^{N-1}\sum _{h=k}^{N-1}y\left[{m}\right]y^{*}\left[h\right]\sum _{n=0}^{N-1}B\left[{n,m}\right]B^{*}\left[{n,h}\right].}
從此結果可發現
y
[
m
]
y
∗
[
h
]
∑
n
=
0
N
−
1
B
[
n
,
m
]
B
∗
[
n
,
h
]
{\displaystyle y\left[{m}\right]y^{*}\left[h\right]\sum _{n=0}^{N-1}B\left[{n,m}\right]B^{*}\left[{n,h}\right]}
不一定為正的,所以無法利用增加基底數來改善平方誤差。