Image Processing Reference
In-Depth Information
=
2
π
μ
m
,f
Ω
f
(
t
m
)
(6.33)
which uses no integrals, this result is a fairly simple tool to find projection coeffi-
cients.
Example 6.2. We illustrate sampling and periodization by Fig. 6.4.
•
In the top, left graph, we have drawn a band-limited signal
f
(
t
). The very def-
inition of
f
as band-limited implies that
F
(
ω
) has a limited extension in the
spectrum, which is drawn in the top, right graph. The width of
F
is
Ω
.
•
In the middle-left graph, we have drawn
f
after discretization. In the middle
right graph, we have drawn the
ω
-domain after discretization. The discretization
in one domain is equivalent to a periodization in the other domain. Also, the
discretization step is inversely proportional to the period of the other domain.
•
In the bottom, left graph we have shown the sampled
f
, when we periodize
F
,
withalargerperiodthantheextensionof
F
.Thisperiodizationof
F
,usingtwice
as large an
Ω
as compared to the extension of
F
, is shown in the bottom, right
graph.
We summarize the results of this section by the
Nyquist theorem
.
Theorem 6.3 (Nyquist).
Let
f
and
F
be a Fourier transform pair. Then sampling of
either of the functions is equivalent to periodization of the other function. To be more
precise, the following two statements are valid:
T
2
,
T
2
•
(Time-limited signals) If
f
is nonzero only in the interval
[
]
, then
F
can
be sampled without loss of information provided that the sampling period is
2
T
or less.
−
Ω
2
,
Ω
2
•
(Band-limited signals) If
F
is nonzero only in the interval
[
]
then
f
can
be sampled without loss of information provided that the sampling period is
2
Ω
or less.
−
When working with band-limited signals, the distance between the samples in
the
t
-domain can be assumed to be equal to 1, for convenience. Then we obtain the
critical frequency domain parameter
Ω
=2
π
, so that the critical frequency in the
theorem yields:
Ω
2
=
π
⇔−
π<ω<π
(6.34)
Ω
2
In signal processing literature,
is often referred to as the Nyquist frequency. Even
its normalized version,
π
, is called the Nyquist frequency, because a scaling of
ω
and
t
-domains is always achievable. The basic interval
−
pi<ω<π
is sometimes
called the
Nyquist period
,orthe
Nyquist block
.