Image Processing Reference
In-Depth Information
For finite sampling, there are not sufficient data to define this series. The
conventional Fourier reconstruction method treats the unknown coefficients as
zero and, as a result, we have
N
/
21
−
∑
1
ˆ
()
ρ
x
=
∆
k
Dn
(
∆
k e
)
i
2
π
n x
∆
, | |
x
<
,
(2.3)
2
∆
k
nN
=−
/
2
which can be evaluated efficiently using the fast Fourier transform (FFT) algo-
rithm. Some basic properties of the Fourier reconstruction method are summa-
rized in the following remarks:
ˆ
()
Remark
1: Given
D
(
n
∆
k
) for
n
=
−
N
/2,
−
N
/2
+
1,
…
,
N
/2
−
1, any
ρ
x
given below with satisfies, Equation 2.1,
N
−
∑
/
21
/;
ˆ
()
ρ
x
=
∆
k
Dn
(
∆
k e
)
i
2
π
n x
∆
+
c e
i
2
π
n
∆
kx
,
(2.4)
n
nN
=−
/
2
nNnN
<−
2
≥
/
2
which is often called a feasible reconstruction of
ρ
(
x
).
ˆ
()
Remark
2: The Fourier reconstruction,
ρ
x
, given in Equation 2.3 is the
minimum-norm feasible solution because
1
∫
2
∆
k
|
ˆ
()|
c
=
arg min
ρ
x
2
dx
=
0
.
(2.5)
n
1
−
2
∆
k
ˆ
()
Remark
3: The Fourier reconstruction,
ρ
x
, is related to the true image
ρ
(
x
) by
1
∫
2
∆
k
ˆ
()
(
ˆ
)(
ˆ
)
ˆ
,
ρ
x
=
ρ
x h x
−
x dx
(2.6)
1
−
2
∆
k
where
h
(
x
), known as the point spread function (PSF), is given by
sin(
π
π
Nkx
kx
∆
∆
)
hx
()
=
∆
k
e
−
i x
π
∆
.
(2.7)
sin(
)
Note that
h
(
x
) is a periodic function, and within each period it is similar to
a sinc function. The width of its main lobe, as measured by the interval between
the first two zero crossings, is 2/(
N
∆
k
). The effective width
W
h
of
h
(
x
) is often
Search WWH ::
Custom Search