Digital Signal Processing Reference
In-Depth Information
approximation
P
−1
f
.
2
(R)
f
L
f
P
-1
f
P
0
f
Q
0
f
t
(b)
(a)
(c)
Figure 2.11
Approximation of (a) P
0
f, (b) P
−1
f, and (c) the detail signal Q
0
f, with
P
0
f+Q
0
f=P
−1
f.
The scaling function coecients for the Haar wavelet at scale
m
are
given by
c
m,n
=
<f,ϕ
mn
>
=2
−m/2
2
m
(n+1)
2
m
n
f
(
t
)
dt
(2.66)
This yields an approximation of
f
at scale
m
:
P
m
f
=
n
c
m,n
ϕ
mn
(
t
)=
n
c
m,n
2
−m/2
ϕ
(2
−m
t
−
n
)
(2.67)
In spite of their simplicity, the Haar wavelets exhibit some undesirable
properties which pose a di
culty in many practical applications. Other
wavelet families, such as Daubechies wavelets and Coiflet basis [4, 278]
are more attractive in practice. Daubechies wavelets are quite often used
in image compression. The scaling function coecients
h
0
(
n
)andthe
wavelet function coecients
h
1
(
n
) for the Daubechies-4 family are nearly
impossible to determine. They were obtained based on iterative methods
[38].
Multiscale Signal Decomposition and Reconstruction
In this section we will illustrate multiscale pyramid decomposition.
Based on a wavelet family, a signal can be decomposed into scaled and
translated copies of a basic function. As discussed in the preceeding
sections, a wavelet family consists of scaling functions, which are scalings
and translations of a father wavelet, and wavelet functions, which are
Search WWH ::
Custom Search