1 a directional extension for multidimensional wavelet ...minhdo/publications/dew.pdf · 3 ω 4 hh...
TRANSCRIPT
1
A Directional Extension for Multidimensional
Wavelet TransformsYue Lu∗ and Minh N. Do
IP EDICS: 2-WAVP (Wavelets and Multiresolution Processing).
Abstract
Directional information is an important and unique feature of multidimensional signals. As a result
of a separable extension from one-dimensional (1-D) bases, multidimensional wavelet transforms have
very limited directionality. Furthermore, different directions are mixed in certain wavelet subbands. In
this paper, we propose a simple directional extension for wavelets (DEW) that fixes this subband mixing
problem and improves the directionality. The building block of the DEW is a two-channel 2-D filter
bank with a checkerboard-shaped frequency partition. The DEW works with both the critically-sampled
wavelet transform as well as the undecimated wavelet transform. In the 2-D case, it further divides
the three wavelet subbands (i.e. horizontal, vertical, and diagonal) at each scale into six finer directional
subbands. The DEW itself is critically-sampled, and hence will not increase the redundancy of the overall
transform. Though nonseparable in essence, the proposed DEW has an efficient implementation that only
requires 1-D filtering. Meanwhile, the DEW can be easily generalized to higher dimensions. In a nutshell,
the proposed directional extension provides an optional tool to efficiently enhance the directionality of
multidimensional wavelet transforms. Numerical experiments show that certain wavelet-based image
processing applications will benefit from this improved directionality.
Index Terms
Wavelet transform, directional information, checkerboard filter bank, filter design, multidimensional
signal processing, image denoising, feature extraction.
Y. Lu is with the Department of Electrical and Computer Engineering and the Coordinated Science Laboratory, University of
Illinois at Urbana-Champaign, Urbana IL 61801 (e-mail: [email protected]; WWW: http://www.ifp.uiuc.edu/∼yuelu).
M. N. Do is with the Department of Electrical and Computer Engineering, the Coordinated Science Laboratory, and
the Beckman Institute, University of Illinois at Urbana-Champaign, Urbana IL 61801 (e-mail: [email protected]; WWW:
http://www.ifp.uiuc.edu/∼minhdo).
This work was supported by the National Science Foundation under Grant CCR-0237633 (CAREER).
April 1, 2005 DRAFT
2
I. INTRODUCTION
Directional information is a unique feature of multidimensional (MD) signals. Recently, the importance
of directional information has been recognized by many image processing applications, including feature
extraction, enhancement, denoising, classification, and compression.
The wavelet transform [1], [2], [3] has a long and successful history as an efficient image processing
tool. However, as a result of a separable extension from one-dimensional (1-D) bases, wavelets in higher
dimensions can only capture very limited directional information. For instance, 2-D wavelets only provide
three directional components, namely horizontal, vertical, and diagonal. Furthermore, the 45◦ and 135◦
directions are mixed in diagonal subbands.
There have been a number of approaches in providing finer directional decomposition. Some notable
examples include 2-D Gabor wavelets [4], the steerable pyramid [5], the directional filter bank [6], 2-D
directional wavelets [7], brushlets [8], complex wavelets [9], [10], [11], curvelets [12], and contourlets
[13], [14]. However, the wavelet transform is still very attractive for image processing applications for
a number of reasons. First, the wavelet transform has a critically-sampled implementation, and can be
easily extended to multidimensional cases. In contrast, most of the approaches mentioned above are
expansive systems with various redundancy ratios. For example, the complex wavelet transform is 4-
times redundant for images, and in general 2N -times redundant for N -dimensional signals. In terms of
implementation complexity, the multidimensional wavelet transforms can be implemented efficiently in
a separable fashion. In contrast, systems such as the directional filter bank involve nonseparable filtering
and sampling and have high computational complexity. Last but not least, the theory and applications
of wavelets have already been extensively studied, offering us a plethora of ready-to-use filters and
processing algorithms.
Therefore the natural question is: “Can we extend the wavelet transform with finer directionality, while
still retain its structure and desirable features?” We give an affirmative answer in this paper by proposing
a simple directional extension for wavelets (DEW). Based on a nonredundant checkerboard filter bank,
the proposed DEW works with both the critically-sampled wavelet transform as well as the undecimated
wavelet transform. In the 2-D case, the DEW leads to one lowpass subband and six directional highpass
subbands at each scale, and fixes the subband mixing problem of wavelets (see Figure 1). Being critically-
sampled itself, the DEW will not increase the redundancy of the overall transform. Though nonseparable
in essence, the DEW has an efficient implementation based on 1-D operations only. Finally, the DEW
can be easily generalized to higher dimensions.
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 3
������������������������������������������������������������������������������������������������������������������������
������������������������������������������������������������������������������������������������������������������������
������������������������������������������
������������������������������������������
������������������������������������������������������������������������������������������������������������������������
������������������������������������������������������������������������������������������������������������������������������������������������������������
������������������������������������
������������������������������������
������������������������������������
������������������������������������������
������������������������������������������
������������������������������������������������������������������������������������������������������������������������
������������������������������������������������������������������������������������������������������������������������
��������������������������������������������������������������������������������������������������������������������������������������������
��������������������������������������������������������������������������������������������������������������������������������������������
HL
HHHH LH
LL HL
LHHH HH
ω1
ω2 (π, π)
−(π, π)
(a)
����������������
����������������
����������������
����������������
����������������
��������������������
������������
����������������
����������������
���������������� ����
������������
����������������
����������������
����������������
������������������������������������
������������������������������������
����������������
����������������
������������������������������������
������������������������������������
������������������������������������
������������������������������������
������������������������������������
������������������������������������
1
1
2
4
34
0
3
56
65
2
ω1
ω2 (π, π)
−(π, π)
(b)
Fig. 1. Division of the 2-D frequency spectrum. (a) The frequency decomposition of the wavelet transform. (b) The frequency
decomposition of the wavelet transform with the proposed directional extension.
The outline of the paper is as follows. Section II presents the filter bank construction of the directional
extension in 2-D, with emphasis on intuition and illustration. The generalization of the results to higher
dimensional cases is given in Section III with a more rigorous treatment. Section IV discusses filter
design and efficient implementation. We will present some numerical results in Section V and conclude
the paper in Section VI. Some preliminary results of this paper were reported in an earlier conference
paper [15].
II. THE DIRECTIONAL EXTENSION IN THE 2-D CASE
The traditional way to construct 2-D wavelets is to use tensor products of their 1-D counterparts.
The advantage of this approach is its simple separable implementation. Unfortunately, this also imposes
serious limits on the directionality of the resulting frequency partitioning. As shown in Figure 1(a), the
2-D wavelet transform produces one lowpass subband (LL), and three highpass subbands (HL, LH, HH),
corresponding to the horizontal, vertical, and diagonal directions. Furthermore, diagonal subbands mixes
the directional information oriented at 45◦ and 135◦. The main idea here is to find some directional
extension to further divide each highpass subband of the wavelets into two branches. In particular,
we want to have a system with the frequency partitioning shown in Figure 1(b), which contains six
directional subbands roughly oriented at 15◦, 45◦, 75◦, 105◦, 135◦ and 165◦. This is the same frequency
decomposition provided by the 2-D dual-tree complex wavelet transform [9], [10], [11], which has been
shown to be successful in several image processing applications. However, the 2-D complex wavelet
transform is 4-times redundant and uses a different filter bank structure compared to wavelets. In the
following, we will discuss our directional extension for wavelets in two cases, i.e., when the wavelet
transform is critically-sampled or undecimated.
April 1, 2005 DRAFT
4
horizontal
vertical
LL
LH
HL
HH
D2
D2
D2
D2
D1
D1
F0(z2)
F0(z2)
F0(z1)
F1(z2)
F1(z2)
F1(z1)
Fig. 2. The filter bank construction of the 2-D DWT for one level of decomposition. The filtering and downsampling operations
are performed in a separable fashion. F0 and F1 denote the 1-D lowpass and highpass filters, with dark regions representing the
ideal passband. D1 = diag(2, 1) and D2 = diag(1, 2) are the 2-fold downsampling matrices along the horizontal and vertical
directions, respectively.
A. Working with the Critically-Sampled Discrete Wavelet Transform (DWT)
This case corresponds to the most widely-used wavelet implementation in which the lowpass and
highpass filters are always followed by a 2-fold downsampling. Figure 2 shows the filter bank imple-
mentation of the DWT for one level of decomposition. For notations used in the figure and hereafter,
we use lower-case letters x[n], where n = (n1, n2)T , to denote 2-D discrete signals or filters. We use
the corresponding uppercase letters X(z1, z2) and X(ejω1 , ejω2) for their z-transforms and discrete-time
Fourier transforms, respectively. When the signal or filter is 1-D, the above notations will be simplified as
x[ni], X(zi) and X(ejωi), with i = 1 or 2 specifying the particular dimension. We use the downsampling
matrices
D1 = diag(2, 1) =
⎛⎝ 2 0
0 1
⎞⎠ and D2 = diag(1, 2) =
⎛⎝ 1 0
0 2
⎞⎠
to represent the 2-fold downsampling operations along the horizontal and vertical directions, respectively.
In general, the multidimensional (MD) downsampling operation [16] is specified by an integer matrix
M as y[n] = x[Mn], where x[n] and y[n] are the MD input and output signals.
To see what the desired directional extension should be, we can examine the frequency contents in the
wavelet subbands. For simplicity, we first consider the case of using ideal filters, where the frequency
response of each filter is constant in the passband and exactly zero in the stopband. We know the
diagonal subband (HH) captures certain directional highpass frequency information (illustrated as regions
{a, b, c, d} in Figure 3(a)) in the input image, where {a, d} and {b, c} correspond to directional information
oriented at 45◦ and 135◦, respectively. With the downsampling operations in the wavelet transform, these
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 5
������������������������������������
������������������������������������
������������������������������������������
������������������������������������������
������������������������������������
������������������������������������
������������������������������������������
������������������������������������������
ω1
ω2 (π, π)
−(π, π)
a b
c d
(a)
ω1
ω2 (π, π)
−(π, π)
ab
cd
(b)
Fig. 3. (a) The diagonal highpass frequency regions of the input signal. (b) The frequency contents of the diagonal subband
(HH) after being downsampled by 2 in each dimension.
H0(z1, z1)
H1(z1, z2)
G0(z1, z2)
G1(z1, z2)
D1
D1 D1
D1
Fig. 4. The two-channel 2-D filter bank with a checkerboard-shaped frequency partition. The dark regions represent the ideal
passband.
frequency regions will be scrambled and mapped to the actual frequency contents in the HH subband, as
shown in Figure 3(b). Now to separate regions {a, d} from {c, b}, we can see that a natural choice is to
use a two-channel 2-D filter bank with a checkerboard-shaped frequency partition, illustrated in Figure 4.
The downsampling matrix used in the filter bank is the simple diagonal matrix D1. 1
Similarly, it can be checked that the same checkerboard filter bank can also be used to divide the
other two wavelets subbands (HL and LH). This suggests that we can use the checkerboard filter bank
as the building block for a directional extension for the DWT to improve its directionality, as shown in
Figure 5, where only the analysis part is given. The original wavelet transform is kept in the first two
levels. In the third level, the three highpass subbands are further split by the checkerboard filter bank.
Applying the multirate identities [16] in multirate signal processing, we can get the equivalent filter of
each subband and verify that the system indeed achieves the desired frequency partitioning (Figure 1(b))
with ideal filters.
Since each individual component of the proposed system, i.e., the DWT and the checkerboard filter
1We can also use D2 as the downsampling matrix.
April 1, 2005 DRAFT
6
2
1
3
4
0LL
HL
HH
LH
separable 2−D wavelet transform directional extension
5
6
D1
D1
D1
D1
D1
D1
WT
Fig. 5. The filter bank construction of the proposed system for one level of decomposition. The system can be iterated on the
lowpass subband 0 for multiple levels of decomposition. The synthesis part, i.e., the inverse transform, is given by a concatenation
of the synthesis part of the checkerboard filter banks and the inverse wavelet transform.
bank, is critically sampled, the overall system is also critically sampled. Furthermore, if we design the
checkerboard filter bank to be perfect reconstruction, then the whole system is also perfect reconstruction.
A nice property of the wavelet transform is that it has an efficient separable implementation. As shown in
Section IV, the proposed directional extension also has an efficient 1-D implementation in the polyphase
domain, and hence will only moderately increase the computational complexity of the wavelet transform.
We recognize that the proposed system has a fundamental limit in its frequency response, when we
use non-ideal filters. Consider the chain of operations in a certain branch, say subband 5, in Figure 5.
D1 D1D2
F1(z1) F1(z2) H0(z1, z2)
Here, F1(z1), F1(z2) are the wavelet highpass filters along the horizontal and vertical directions, and
H0(z1, z2) is one of the checkerboard filters. Using the multirate identities, we can rewrite this branch
in its equivalent form
F1(z1)F1(z2)H0(z21 , z2
2) M
where the equivalent downsampling matrix is M = D1 · D2 · D1 = diag(4, 2).
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 7
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
0
1
2
(a) |F1(ejω1)F1(e
jω2)|
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
0
0.5
1
(b) |H0(ej2ω1 , ej2ω2)|
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
0
1
2
(c) |Feq(ejω1 , ejω2)|
Fig. 6. The magnitude frequency responses of F1(ejω1)F1(e
jω2), H0(ej2ω1 , ej2ω2) and Feq(e
jω1 , ejω2) for the DWT case.
In the Fourier domain, the equivalent filter of this branch is
Feq(ejω1 , ejω2) = F1(ejω1)F1(ejω2)H0(ej2ω1 , ej2ω2). (1)
In Figure 6(a) and Figure 6(b), we show the typical magnitude frequency responses of some non-ideal
F1(ejω1)F1(ejω2) and H0(ej2ω1 , ej2ω2). The magnitude frequency response of the equivalent filter Feq is
shown in Figure 6(c) as the multiplication of the two previous responses. We can see that |Feq(ejω1 , ejω2)|is concentrated mostly along the 135◦ direction. However, there are also some unwanted bumps (aliasing
components) along the 45◦ direction. A similar phenomenon was mentioned in [11] for a nonredundant
construction of the complex wavelet transform. Similar to the observations made there, it is only possible
to reduce the area of those bumps by using longer filters with sharper cut-off behavior; however, the height
of the bumps cannot be reduced. It is expected that this aliasing problem will have a negative impact
on the performance of the proposed system. However, as shown in Section V, the overall system still
benefits from the improved directionality in some applications, since the equivalent frequency responses
are still mostly concentrated along single directions.
B. Working with the Undecimated Wavelet Transform (UWT)
Another useful wavelet transform is called the undecimated wavelet transform [17], [2], which discards
all the downsampling operations in the DWT and use appropriately upsampled filters via the “algorithme a
trous” [17]. With J denoting the number of decomposition levels, the UWT is a (3J +1)-times redundant
transform in 2-D. Note that the lowpass/highpass filters in coarser levels are upsampled accordingly so
that the overall equivalent filters are the same as those from the DWT. Similar to the discussion above, we
can improve the directionality of the UWT by attaching checkerboard filter banks (directional extension)
to its highpass branches. The construction is similar to the one shown in Figure 5, with the DWT replaced
April 1, 2005 DRAFT
8
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
0
0.5
1
(a) |H0(ejω1 , ejω2)|
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
0
1
2
(b) |Feq(ejω1 , ejω2)|
Fig. 7. The magnitude frequency responses of H0(ejω1 , ejω2) and Feq(e
jω1 , ejω2) for the UWT case.
by the UWT in the wavelet part. Since the checkerboard filter banks are still downsampled, the directional
extension will not increase the redundancy of the overall system.2
For the UWT case, the frequency aliasing problem is eliminated. For example, the chain of operations
in subband 5 of Figure 5 will now be
D1
F1(z2)F1(z1) H0(z1, z2)
The Fourier transform of the equivalent filter is
Feq(ejω1 , ejω2) = F1(ejω1)F1(ejω2)H0(ejω1 , ejω2).
Compared with (1), here we are using H0(ejω1 , ejω2) (Figure 7(a)) instead of its upsampled version
H0(ej2ω1 , ej2ω2) (Figure 6(b)) to divide the wavelet subband. This makes a big difference when we use
filters with non-ideal frequency responses. As shown in Figure 7(b), the frequency response Feq(ejω1 , ejω2)
for the UWT case is single directional and free of the unwanted aliasing components.
C. Connections with Other Related Systems
As mentioned before, the 2-D wavelet transform with the proposed DEW achieves the same frequency
partitioning as that from the 2-D dual-tree complex wavelet transform (CWT). However, there are several
differences that are important to mention. The CWT produces complex coefficients and hence provides
the phase information, which has be shown to be very useful for some signal processing applications.
2However, the overall system is no longer shift-invariant due to the downsampling.
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 9
As a price to pay for that, the CWT is an expansive system, being 4-times redundant for images, and in
general 2N -times redundant for N -dimensional signals. In contrast, the proposed DEW is a real-valued
transform, and hence discards the phase information. However, it can be nonredundant (when working
with the DWT) for arbitrary dimensions. More importantly, the scaling filters in the two trees of the
CWT have to satisfy the half-sample delay condition, and hence cannot be arbitrarily chosen. This leads
to new challenging joint filter design problems. However, our proposed DEW can work with any existing
wavelet filters to suit different applications. No filter design on the wavelet part is needed. Actually this
difference reflects the philosophy of this work: we try to improve the directionality of wavelets in an
efficient and unintrusive way. Any existing software or hardware implementation of the wavelet transform
can continue to be used. Just attach the directional extension!
In [18], [19], Fernandes et al. proposed a new non-redundant version of the complex wavelet transform
through either a pre-mapping or a post-mapping step. In that interesting construction, the phase informa-
tion can be obtained without sacrificing in redundancy. One major difference between the two approaches
is that the inverse pre-projection and post-projection steps need to use infinite impulse response (IIR)
filters to achieve perfect reconstruction. In contrast, our proposed system has perfect reconstruction with
linear-phase finite impulse response (FIR) filters.
III. GENERALIZATION TO HIGHER DIMENSIONS
In this section, we generalize the proposed directional extension for wavelets to arbitrary dimensions.
Let us first take a look at the 3-D case. We know the 3-D discrete wavelet transform gives one lowpass
subband and seven highpass subbands. Just like its 2-D counterpart, it suffers from limited directionality
and the subband mixing problems. Figure 8 shows that the situation here is actually even more severe.
Analog to the 2-D case, our idea is to find a DEW to improve the directional selectivity of these highpass
subbands.
A. Partitioning the N -Dimensional Frequency Spectrum
We start with some notations. Let B denote the set of frequencies in [−π, π). The frequency spectrum
of N -dimensional (N -D) discrete signals is the Cartesian product BN = B × B × . . . × B︸ ︷︷ ︸N
. Let B+0 =
[0, π/2), B−0 = [−π/2, 0), B+
1 = [π/2, π), and B−1 = [−π,−π/2). One can easily verify that the
one-level separable N -D wavelet transform (both the DWT and the UWT) decomposes BN into 2N
April 1, 2005 DRAFT
10
(a) (b)
Fig. 8. Frequency-domain support of two 3D-DWT subbands. Dark regions represent the ideal passband. In the extreme case,
information corresponding to four completely different directions is mixed in subband (a).
subbands
BN =⋃
ki∈{0,1}B±
k1× B±
k2× . . . × B±
kN
def=⋃
ki∈{0,1}B±
k1B±
k2. . . B±
kN, (2)
where B±ki
def= B+ki∪ B−
kifor i = 1, . . . , N . In other words, if we calculate the equivalent filter 3 of each
wavelet subband, the frequency-domain support of that filter will be one of the 2N regions in (2). For
each highpass subband B±k1
B±k2
. . . B±kN
with at least one ki = 1, we can expand it using the distributive
law
B±k1
B±k2
. . . B±kN
def= (B+k1
∪ B−k1
)(B+k2
∪ B−k2
) . . . (B+kN
∪ B−kN
)
=⋃
si∈{+,−}Bs1
k1Bs2
k2. . . BsN
kN(3)
=⋃
si∈{+,−}
(Bs1
k1Bs2
k2. . . BsN
kN∪ B−s1
k1B−s2
k2. . . B−sN
kN
). (4)
We group the 2N terms in (3) into 2N−1 different polarized pairs in (4), so that each of them corresponds
to the frequency support of some real-valued filter. For the 3-D case, we show in Figure 9(a) and
Figure 9(b) two possible subbands from (4). We can see that each of them corresponds to a single
direction, which is in sharp contrast to the situation in Figure 8(a). In addition to the “full” splitting in
(4), it is also possible to have other “partial” splitting. For example,
B±k1
B±k2
. . . B±kN
=⋃
si∈{+,−}
(Bs1
k1. . . B
sn+1
kn+1∪ B−s1
k1. . . B
−sn+1
kn+1
)B±
kn+2. . . B±
kN, (5)
where n = 1, . . . , N − 1. The “full” splitting is a special case when n = N − 1. Figure 9(c) shows
one of the possible subbands from the “partial” splitting in 3-D. These subbands are useful, since they
3In this section we only consider ideal filters.
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 11
(a) (b) (c)
Fig. 9. Frequency-domain support of several subbands, with dark regions representing the ideal passbands. Part (a) and Part
(b): Two possible subbands from the “full” splitting in (4). Each of them corresponds to a single direction, which is in sharp
contrast to the situation in Figure 8(a). Part (c): One possible subband from the “partial” splitting in (5).
correspond to filters that can capture signals with singularities lying on some low-dimensional manifold
structures (e.g. curves, surfaces, etc.) [20].
B. The Directional Extension for Wavelets in N -D
The main result in this section is that we can achieve the subband splitting in (4) and (5) by attaching
some directional extension to the highpass subbands of the wavelet transform. The basic building blocks
of the N -D DEW are checkerboard filter banks shown in Figure 10(a). For notations, we use CBi,j
to denote a checkerboard filter bank whose filtering operations are carried out along the ith and jth
dimensions. The downsampling matrix is defined to be Di = diag(1, . . . , 1︸ ︷︷ ︸i−1
, 2, 1, . . . , 1), which keeps
every other sample along the ith dimension. The two channels of the (analysis) filter bank are denoted
by CBi,j(0) and CBi,j(1), respectively.
Lemma 1: Suppose we attach the filter bank CBi,i+1 (1 ≤ i ≤ N − 1) to one of the highpass
subbands of the N -D wavelet transform (either the DWT or the UWT) with an ideal frequency sup-
port B±k1
B±k2
. . . B±kN
. If all filters used are ideal, the wavelet subband will be split into two channels
with passband support B±k1
. . . B±ki−1
(B+ki
B+ki+1
∪ B−ki
B−ki+1
)B±ki+2
. . . B±kN
and B±k1
. . . B±ki−1
(B+ki
B−ki+1
∪B−
kiB+
ki+1)B±
ki+2. . . B±
kN, respectively.
Remark 1: This is the generalization of the 2-D result in Section II and is proved by applying the
multirate identities. See Appendix for details.
As shown in Figure 10(b), an n-level DEW is a binary tree-structured expansion of filter banks
CB1,2, CB2,3, . . ., and CBn,n+1, for which we have the following theorem:
Theorem 1: Suppose we attach the n-level DEW (1 ≤ n ≤ N −1) to one of the highpass subbands of
the N -D wavelet transform (either the DWT or the UWT) with an ideal frequency support B±k1
B±k2
. . . B±kN
.
April 1, 2005 DRAFT
12
1
0
H0(zi, zj)
H1(zi, zj)
CBi,j
Di
Di
(a)
00
0
0
0
1
1
1
1
1
CB1,2
CB2,3
CB2,3
CBn,n+1
CBn,n+1
2n
subbands
n-level DEW
(b)
Fig. 10. (a) CBi,j is the analysis part of the checkerboard filter bank. The subscripts i, j (i �= j) specify that the filters H0(zi, zj)
and H1(zi, zj) operate along the ith and jth dimensions. (b) The n-level DEW is a binary tree-structured concatenation of
checkerboard filter banks CB1,2, CB2,3, . . ., and CBn,n+1. Notice the arrangement of subscripts for the simple expansion rule.
If all filters used are ideal, the wavelet subband will be split into 2n channels as in (5).
Remark 2: This is proved by successively applying the result in Lemma 1. See Appendix for details.
Note that the “partial” splitting in (5) is along dimensions 1, 2, . . . , n+1. However, through a permutation
of indices, we can construct a similar n-level DEW to generate “partial” splitting along n + 1 arbitrary
dimensions d1, d2, . . . , dn+1, where di ∈ {1, . . . , N}. Moreover, the “full” splitting in (4) is a special
case when n = N − 1.
In some applications, e.g., enhancing a particular direction in the signal, one might be only interested
in getting a single frequency region. Instead of using the full binary tree expansion of 2n−1 checkerboard
filter banks in Figure 10(b) to get all 2n subbands, we can use a connected sequence of n checkerboard
filter banks for that specific subband. Such a sequence of checkerboard filter banks is specified by the
following proposition, which can be verified in the proof of Theorem 1 (Appendix B).
Proposition 1: Suppose the subband of interest is(Bs1
k1. . . B
sn+1
kn+1∪ B−s1
k1. . . B
−sn+1
kn+1
)B±
kn+2. . . B±
kN,
where si ∈ {+,−}. The desired “adaptive” DEW for that subband will be
CB1,2(c1) → CB2,3(c2) → . . . → CBn,n+1(cn),
where CBi,i+1 is connected to channel # ci−1 (either 0 or 1) of CBi−1,i (i = 2, . . . , n) with the following
expansion rule:
for DWT:
ci =
⎧⎪⎨⎪⎩
1 − δ(ki − ki+1), if si �= si+1
δ(ki − ki+1), if si = si+1
;
for UWT:
ci =
⎧⎪⎨⎪⎩
0, if si �= si+1
1, if si = si+1
,
where i = 1, . . . , n, and δ(·) is the Kronecker delta function.
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 13
H0(z)
H1(z)
G0(z)
G1(z)D1D1
D1 D1
x x
y0
y1
Fig. 11. A maximally decimated two-channel 2-D filter bank with analysis filters H0(z) (zdef= (z1, z2)), H1(z), synthesis
filters G0(z), G1(z), and the downsampling matrix D1. x and x are the input and reconstructed signals, respectively. y0 and
y1 are the two output signals of the analysis part.
H(z) G(z)z−11
z1
D1
D1
D1
D1x x
y0
y1
Fig. 12. The equivalent polyphase form of the filter bank in Figure 11. H(z) and G(z) are 2-by-2 polyphase matrices.
IV. FILTER DESIGN AND EFFICIENT IMPLEMENTATION
As stated before, there is no need to modify the wavelet transform part and we can choose any existing
wavelet filters to suit different applications. The remaining task is to design a 2-D checkerboard filter
bank with the desired frequency response and perfect reconstruction. Here we will present a design
based on a parameterization of the polyphase matrices. The advantage of this design is that the resulting
filter banks have efficient implementation that only requires 1-D filtering. A more general design of
checkerboard filter banks, in particular the ones that are tailored to work in combination with wavelets,
will be investigated in a future work.
A. Design by Polyphase Matrices Parameterization
Let us consider a maximally decimated two-channel filter bank in 2-D with the downsampling matrix
D1, shown in Figure 11. The well-known equivalent polyphase form of the system is given in Figure 12.
To simplify notation, we denote z = (z1, z2). The relationship between the analysis and synthesis filters
{Hk(z), Gk(z)} and the polyphase matrices H(z) and G(z) can be expressed as [16]
Hk(z) = Hk,0(z21 , z2) + z−1
1 Hk,1(z21 , z2) (6)
Gk(z) = G0,k(z21 , z2) + z1 G1,k(z2
1 , z2),
April 1, 2005 DRAFT
14
for k = 0, 1. We can see from Figure 12 that a sufficient condition for the filter bank to have perfect
reconstruction is
H(z) · G(z) = I, (7)
where I is the identity matrix. In our design, we choose H(z) and G(z) to be
H(z) =√
2
⎛⎝ 0.5 0
−0.5α(z) 1
⎞⎠⎛⎝ 1 α(z)
0 z1
⎞⎠ (8)
and
G(z) =1√2
⎛⎝ 1 −z−1
1 α(z)
0 z−11
⎞⎠⎛⎝ 2 0
α(z) 1
⎞⎠ ,
where α(z) is a free design parameter. It is easy to verify that the perfect reconstruction condition (7)
is structurally guaranteed for arbitrary choice of α(z). This form of polyphase parameterization was
first proposed by Phoong et al. [21] for the 1-D filter bank. However, the 1-D to 2-D mapping method
proposed in that work can only be used to design 2-D filters banks with a single parallelogram-shaped
support, such as the diamond shape. While the checkerboard shape we want here does not belong to this
class. In the following, we will propose a novel 1-D to 2-D mapping for the checkerboard filter bank in
the polyphase domain.
Substituting (8) into (6), we get
H0(z) =1 + z−1
1 α(z21 , z2)√
2(9)
Similarly, we can also write down H1(z), G0(z) and G1(z). Actually, they are all related to H0(z) as
follows.
H1(z) = z1
(√2 −
(√2H0(z) − 1
)H0(z)
)(10)
G0(z) = −z−11 H1(−z1, z2) (11)
G1(z) = z−11 H0(−z1, z2). (12)
If H0(z) is the ideal filter with the desired checkerboard-shaped frequency support shown in Figure 4,
i.e., if its Fourier transform takes the constant value√
2 in the passband and 0 in the stopband, we can
then verify from (10) - (12) that the other three filters H1(z), G0(z) and G1(z) will also achieve the
desired frequency response. For instance, G1(ejω1 , ejω2) = e−jω1H0(ej(ω1−π), ejω2), i.e., G1(ejω1 , ejω2)
is obtained by shifting H0(ejω1 , ejω2) horizontally by π and multiplying a linear phase.
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 15
ω1ω1 ω1
ω2 ω2ω2(π, π)(π, π) (π, π)
−(π, π) −(π, π) −(π, π)
+j+j
+j
+j
−j−j
−j
−j
+1
+1
−1
−1
M(z1) M(z2)z−11 α(z2
1 , z2)
Fig. 13. The separable decomposition of z−11 α(z2
1 , z2). The values of the 2-D Fourier transform of the filters are shown in the
figure.
Therefore, we only need to design H0(z) to approximate the ideal filter on the checkerboard support.
In turn, this implies that the Fourier transform of the filter z−11 α(z2
1 , z2) should take constant values
(−1, 1,−1, 1) in the four quadrants of the 2-D frequency plane, as illustrated in Figure 13. Since this is
a separable filter, we can decompose it as the product of two 1-D filters M(z1) and M(z2), i.e.,
z−11 α(z2
1 , z2) = M(z1) · M(z2). (13)
If we further constrain M(z) to have real coefficients, i.e., M(ejω) = M∗(e−jω), then one of the only
two choices is 4
M(ejω) =
⎧⎪⎨⎪⎩−j, for ω ∈ (0, π]
+j, for ω ∈ (−π, 0]. (14)
Meanwhile, the decomposition form in (13) also implies that
M(z) = z−1β(−z2), (15)
for some 1-D filter β(z), and
α(z) = z−12 β(−z1)β(−z2
2). (16)
From (14) and (15), we want β(z) to be an allpass filter with half-sample delay specified as:
|β(ejω)| = 1, ∀ω (17)
∠β(ejω) = 0.5ω, ∀ω ∈ (−π, π); (18)
In summary, the proposed filter bank design process is given as follows.
(1) Design a filter β(z) to approximate the conditions in (17) and (18).
4The only other choice is to use −M(ejω).
April 1, 2005 DRAFT
16
TABLE I
THE COEFFICIENTS OF THE FILTERS β(z) WITH d = 3, 4, 5, 6.
b(n) d = 3 d = 4 d = 5 d = 6
b(0) 0.6141 0.6217 0.6266 0.6300
b(1) -0.1515 -0.1705 -0.1835 -0.1929
b(2) 0.0473 0.0670 0.0839 0.0972
b(3) -0.0240 -0.0383 -0.0526
b(4) 0.0161 0.0272
b(5) -0.0144
(2) Obtain the analysis and synthesis filters H0, H1, G0, G1 from (16) and (9) - (12).
Note that step (2) is only needed for filter bank analysis. As we will see later, knowing β(z) is sufficient
for actual implementation (see Figure 16).
B. Designing the Allpass Filter with Half-Sample Delay
The half-sample delay conditions (17) and (18) can only be realized by an IIR filter with an irrational
system function. There are various ways to design FIR filters to approximate the conditions. See [22]
for a review. Here we use a simple approach by choosing β(z) to be a type-II linear phase filter [23] of
length-2d:
β(z) = b0(1 + z) + b1(z−1 + z2) + . . . + bd−1(z−(d−1) + zd).
Note that the phase condition (18) is satisfied exactly, and we need to design the coefficients b0, . . . , bd−1
such that the magnitude of the frequency response
|β(ejω)| =d−1∑k=0
2bk · cos(
(k +12)ω)
is as close to 1 as possible.
In Table I, we list the coefficients of the filters β(z) of different lengths designed by the Parks-
McClellan algorithm. We show in Figure 14 the magnitude responses of those β(z). From (16), (9),
and (10), we can verify that the resulting 2-D filters H0 and H1 are of sizes (4d − 1) × (4d − 1) and
(8d− 3)× (8d− 3), respectively. In Figure 15, we give the magnitude response of the 2-D filter H0(z)
obtained from a β(z) with d = 6. We can see that it is a good approximation to the ideal checkerboard
filter.
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 17
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9−70
−60
−50
−40
−30
−20
−10
0
10
Normalized Frequency (×π rad/sample)
Mag
nitu
de (
dB)
Magnitude Response (dB)
d = 3
d = 4
d = 5
d = 6
Fig. 14. The magnitude frequency responses of β(z)
with d = 3, 4, 5, and 6. Note that the phase responses
(not shown in the figure) satisfy (18) exactly.
−1
−0.5
0
0.5
1
−1
−0.5
0
0.5
1
0
0.5
1
1.5
Fx
Fy
Mag
nitu
de
Fig. 15. The magnitude frequency response of the
filter H0(z) obtained from a β(z) with d = 6.
z−12 β(−z1)β(−z2
2) z−12 β(−z1)β(−z2
2)−z−12 β(−z1)β(−z2
2) −z−12 β(−z1)β(−z2
2)
√2z1
1√2z−11
z1z−11
1√2
√2
H(z) G(z)
D1
D1 D1
D1
x xy0
y1
Fig. 16. The efficient polyphase implementation of the 2-D checkerboard filter bank.
C. Efficient Implementation
A nice property of the proposed filter design is that the structure is similar to a ladder network [21].
In Figure 16, we show the polyphase implementation of the analysis and synthesis parts of the filter
bank. Although the designed 2-D filters are nonseparable, their polyphase components are separable.
Meanwhile, the downsampling matrix D1 = diag(2, 1) is also separable. Therefore, the entire system
can be implemented by efficient 1-D operations only.
To have a rough estimate of the computational complexity of the proposed DEW, we can count the
number of arithmetic operations needed for each new input sample to the DEW. In 2-D, the DEW is
just a single-level checkerboard filter bank implemented in the polyphase domain as in Figure 16. For
simplicity, we assume the convolution operations are implemented as straight polynomial products, while
similar analysis can be done for FFT-based convolutions. Suppose the filter β(z) used in Figure 16 is of
April 1, 2005 DRAFT
18
TABLE II
NUMBER OF ARITHMETIC OPERATIONS/INPUT SAMPLE FOR DIFFERENT SYSTEMS. BOTH THE DWT AND THE UWT
PERFORM A J -LEVEL DECOMPOSITION OF A N -DIMENSIONAL SIGNAL. THE TWO 1-D FILTERS USED IN THE WAVELET
TRANSFORM ARE OF LENGTH L1 AND L2 . THE CHECKERBOARD FILTERS USED IN THE DEW ARE DERIVED FROM A β(z) OF
LENGTH 2d.
# of multiplications/input sample # of additions/input sample
DWT (L1+L22
)N · 1−2−NJ
1−2−N (L1+L22
− 1)N · 1−2−NJ
1−2−N
DEW for DWT (2d + 1)(N − 1)(1 − 2−NJ) (4d − 1)(N − 1)(1 − 2−NJ)
UWT (L1 + L2)(2N − 1)J (L1 + L2 − 2)(2N − 1)J
DEW for UWT (2d + 1)(2N − 1)J (4d − 1)(2N − 1)J
length 2d. Since it is a symmetric filter, the filtering by β(−z1) requires d multiplications/input sample
and 2d − 1 additions/input sample. Moreover, the filtering by β(−z22) can be efficiently implemented in
the polyphase domain and still requires d multiplications/input sample and 2d−1 additions/input sample.
We can verify that the total number of operations needed for the analysis part of the DEW is 2d + 1
multiplications/input sample and 4d − 1 additions/input sample. The number of operations needed for
the synthesis part is the same, since the analysis and synthesis parts are duals of each other. For the
general N -dimensional case, the DEW (for “full” splitting) is a (N − 1)-level binary tree expansion
of checkerboard filter banks. The number of operations becomes (2d + 1)(N − 1) multiplications/input
sample and (4d − 1)(N − 1) additions/input sample.
For comparison purposes, we show in Table II the number of arithmetic operations/input sample for
the DWT, the UWT, and the corresponding DEW in either case. We assume both wavelet transforms are
implemented in the polyphase domain 5 and perform a J-level decomposition of a N -dimensional signal.
The lowpass and highpass filters used in the wavelet transforms are of length L1 and L2, respectively. We
do not assume special structures, such as linear phase or orthogonality, on them. The DEW is attached
to all the highpass subbands of the wavelet decomposition tree. Here we just list the numbers without
derivation. A detailed analysis on the computational complexity of multirate systems can be found in [2].
We can see from Table II that the relative increase in computational complexity (by attaching the DEW) is
mainly determined by the ratio d/(L1+L2). In 2-D and for some typical numbers L1 = 9, L2 = 7, d = 6,
5This is especially important for the computational efficiency of the UWT.
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 19
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
0
2
(a) DWT: frequency
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
0
5
(b) DWT-DEW: frequency
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
0
0.2
(c) UWT-DEW: frequency
(d) DWT: spatial (e) DWT-DEW: spatial (f) UWT-DEW: spatial
Fig. 17. A few basis images from different transforms. (a) - (c): Some basis images in the frequency domain. (d) - (f): Some
basis images in the spatial domain.
applying the DEW to the DWT corresponds to an increase of 61% multiplications and 123% additions.
Similarly, applying the DEW to the UWT brings an increase of 81% multiplications and 164% additions.
V. NUMERICAL RESULTS
All experiments in this section use the checkerboard filters designed in Section IV with a β(z) of
length 12 (i.e. d = 6). These checkerboard filters are close to having the ideal frequency response and
thus can provide better separation of the directional information. Meanwhile, the sharp cut-off of these
filters can help alleviate the aliasing problem mentioned in Section II. For the wavelet transform, we use
three levels of decomposition and employ Daubechies’s symmlet of length 12 [1]. As before, we use
DWT and UWT to represent the critically-sampled and the undecimated wavelet transforms, respectively.
Correspondingly, we use DWT-DEW and UWT-DEW to denote the two wavelet transforms equipped
with the proposed directional extension.
A. Basis Images
Figure 17 shows a few basis images from different transforms in both frequency and spatial domains.
The basis images of UWT are not shown here, since they are equivalent to those of DWT. We can
see that the basis images of DWT-DEW and UWT-DEW have better directionality than those of DWT.
April 1, 2005 DRAFT
20
(a) (b)
(c) (d)
Fig. 18. (a) A synthetic image consisting of diagonal lines. (b) The diagonal subband of the DWT. (c) Subband 5 of the
DWT-DEW. (d) Subband 6 of the DWT-DEW.
As expected, the critically-sampled DWT-DEW suffers from the frequency aliasing problem, resulting
in some unwanted bumps in the opposite direction (Figure 17(b)). Although the height of these bumps
cannot be reduced, it is possible to make the area of the bumps small by using filters with sharp cut-
off. However, we should see that most of the subband energy of DWT-DEW concentrates on a single
direction, and hence it might still be useful for some image processing applications. Figure 17(c) shows
that the aliasing problem is eliminated for UWT-DEW, but at the cost of a redundant system.
In the spatial domain, the basis image of DWT from the diagonal (HH) subband exhibits an ap-
parent “checkerboard” artifact (Figure 17(d)). In contrast, DWT-DEW (Figure 17(e)) and UWT-DEW
(Figure 17(f)) succeed in isolating different directions, and no “checkerboard” effect appears.
B. Directional Feature Extraction
We first apply the DWT and DWT-DEW on a synthetic image (Figure 18(a)), which consists of lines
oriented at both diagonal directions. The result of the UWT-DEW is not shown here, since it is very close
to that of the DWT-DEW. Figure 18(b) shows the diagonal (HH) subband of the DWT. As discussed
before, 45◦ and 135◦ directions are mixed in this subband and the separable wavelets cannot discriminate
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 21
(a) (b)
(c) (d)
Fig. 19. (a) The cameraman image. (b) The diagonal subband of the DWT. (c) Subband 5 of the DWT-DEW. (d) Subband 6
of the DWT-DEW.
between them. In Figure 18(c) and Figure 18(d), we show the subbands 5 and 6 of the DWT-DEW.
Clearly the subband mixing problem is solved, and the two subbands correctly capture the corresponding
directional information.
We then do the same experiment on the cameraman image. As shown in Figure 19(c) and Figure 19(d),
the DWT-DEW correctly captures and separates the directional information at 45◦ and 135◦ in two
subbands.
C. Denoising
The improvement in directionality of the transform will lead to improvements in many applications.
As an example, we compare the performance of the DWT and the UWT, both with and without the
proposed DEW, in image denoising. In this experiment, we apply additive white Gaussian noise of
different variances σ2 to the Barbara image. A simple hard thresholding scheme [24] with threshold
T = 3σ is used for all 4 transforms. In Figure 20, we show the peak signal-to-noise ratio (PSNR) of
denoised Barbara images using different transforms under various noise levels. We can see that in all
cases the DEW help improve the performance of the original transforms. For a noisy image with PSNR
April 1, 2005 DRAFT
22
16 17 18 19 20 21 22 23 24 2522
23
24
25
26
27
28
29
30
31
PSNR (dB) of Noisy Images
PS
NR
(dB
) of
Den
oise
d Im
ages
UWT−DEWUWTDWT−DEWDWT
Fig. 20. PSNR (dB) of denoised Barbara images by using different transforms under various noise levels.
= 20.19 dB, the improvement (by using the proposed DEW) is about 0.30 dB for DWT, and a more
significant 1.01 dB for UWT.
Figure 21 displays a “zoom-in” comparison of the denoising results. We can see that edges and
directional textures are better restored when we apply the proposed DEW to the DWT and UWT. Despite
the relatively small improvement in PSNR, the DWT-DEW clearly outperforms the DWT (especially in
regions with directional textures) in terms of visual quality.
VI. CONCLUSION
In this work, we proposed a simple directional extension for wavelets that equips the multidimensional
wavelet transform (both the DWT and the UWT) with finer directionality. The filter bank construction of
the overall system is a concatenation of the separable wavelet transform with a sequence of checkerboard
filter banks. No modification on the wavelet part is necessary and hence any existing wavelet implemen-
tation can continue to be used. In 2-D, the proposed DEW further divides the three wavelet subbands
(i.e. horizontal, vertical, and diagonal) at each scale into six finer directional subbands. The DEW is
nonredundant, and can be easily generalized to higher dimensions. Moreover, the DEW can be optional
and adaptive to specific directions. We also proposed a design of the checkerboard filter banks based
on a parameterization of the polyphase matrices and a novel 1-D to 2-D mapping. The resulting filter
banks have efficient implementation that only requires 1-D filtering. Numerical experiments indicate the
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 23
(a) Original image (b) Noisy image: PSNR = 20.19 dB (c) DWT: PSNR = 24.84 dB
(d) DWT-DEW: PSNR = 25.14 dB (e) UWT: PSNR = 26.72 dB (f) UWT-DEW: PSNR = 27.73 dB
Fig. 21. Comparison of denoising results when applying DWT, DWT-DEW, UWT, and UWT-DEW on the Barbara image.
potential of the proposed DEW in several wavelet-based image processing applications.
APPENDIX
A. Proof of Lemma 1
We assume the wavelet transform is the DWT. The proof for the case of UWT is similar. Suppose we
attach CBi,i+1 to one of the wavelet highpass subbands with an ideal frequency support B±k1
B±k2
. . . B±kN
.
The chain of operations from the input to channel #0 of CBi,i+1 is
→ Fk1(z1) → ↓ D1 → . . . → FkN(zN ) → ↓ DN︸ ︷︷ ︸
N -D DWT
−→ H0(zi, zi+1) → ↓ Di︸ ︷︷ ︸CBi,i+1
→,
April 1, 2005 DRAFT
24
ωi
ωi+1(π, π)
−(π, π)
Fig. 22. The frequency-support of H0(ej2ωi , ej2ωi+1). Dark regions represent the passband.
where Fkjwith kj ∈ {0, 1} are the ideal lowpass and highpass wavelet filters, and H0 is one of the
checkerboard filters in Figure 10(a). Applying the multirate identities, we can rewrite this channel as
→ Fk1(z1) . . . FkN(zN ) · H0(z2
i , z2i+1)︸ ︷︷ ︸
equivalent filter Feq(z1, . . . , zN )
−→ ↓ D1 . . .DN · Di︸ ︷︷ ︸equivalent downsampling matrix
→ .
The frequency-domain support of the equivalent filter is
supp (Feq) = supp(Fk1(e
jω1) . . . FkN(ejωN ) · H0(ej2ωi , ej2ωi+1)
)= supp
(Fk1(e
jω1) . . . FkN(ejωN )
) ∩ supp(H0(ej2ωi , ej2ωi+1)
)= B±
k1B±
k2. . . B±
kN∩ supp
(H0(ej2ωi , ej2ωi+1)
). (19)
The frequency support of H0(ej2ωi , ej2ωi+1) is shown in Figure 22. One can easily verify that
supp (Feq) =
⎧⎪⎨⎪⎩
B±k1
. . . B±ki−1
(B+ki
B+ki+1
∪ B−ki
B−ki+1
)B±ki+2
. . . B±kN
, if ki �= ki+1
B±k1
. . . B±ki−1
(B+ki
B−ki+1
∪ B−ki
B+ki+1
)B±ki+2
. . . B±kN
, if ki = ki+1.
(20)
In either case, the two channels of CBi,i+1 split the wavelet subband into two subbands as described in
Lemma 1.
B. Proof of Theorem 1
Again, we assume the wavelet transform is the DWT. The proof for the case of UWT is similar. The
n-level DEW has 2n subband channels. The chain of operations from the input to any one of the channels
is
→ N -D DWT −→ Hc1(z1, z2) → ↓ D1 → . . . → Hcn(zn, zn+1) → ↓ Dn︸ ︷︷ ︸
one channel of the n-level DEW
→,
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 25
where ci ∈ {0, 1}. Applying the multirate identities, we can get the frequency-domain support of the
equivalent filter as
supp(Fn
eq
)=(B±
k1B±
k2. . . B±
kN
) ∩ supp(Hc1(e
j2ω1 , ej2ω2) . . . Hcn(ej2ωn , ej2ωn+1)
)=(B±
k1B±
k2. . . B±
kN
) ∩(
n⋂i=1
supp(Hci
(ej2ωi , ej2ωi+1)))
=n⋂
i=1
(B±
k1B±
k2. . . B±
kN∩ supp
(Hci
(ej2ωi , ej2ωi+1)))
(21)
= supp(Fn−1
eq
)⋂(B±
k1B±
k2. . . B±
kN∩ supp
(Hcn
(ej2ωn , ej2ωn+1)))
, (22)
where F ieq (i = 1, . . . , n) is the equivalent filter from an i-level DEW. Note that each term in (21) is
the frequency support of a subband produced by attaching a single-level DEW (CBi,i+1) to the wavelet
subband (e.g. see (19)). This implies that we can successively apply the one-step splitting result in
Lemma 1. More formally, we will use an induction proof.
(1) The claim in Theorem 1 is true for n = 1, as ensured by Lemma 1.
(2) Now by induction, we suppose that the (n − 1)-level DEW (2 ≤ n ≤ N − 1) can split the wavelet
highpass subband into 2n−1 regions as⋃si∈{+,−}
(Bs1
k1. . . Bsn
kn∪ B−s1
k1. . . B−sn
kn
)B±
kn+1. . . B±
kN, (23)
and we want to check if the splitting result works for the n-level DEW.
(3) From (22), (19), and (20), each subband in (23) will be split into two in the n-level DEW:((Bs1
k1. . . Bsn
kn∪ B−s1
k1. . . B−sn
kn
)B±
kn+1. . . B±
kN
)⋂(· · · (B+
knB+
kn+1∪ B−
knB−
kn+1) · · ·
),
and ((Bs1
k1. . . Bsn
kn∪ B−s1
k1. . . B−sn
kn
)B±
kn+1. . . B±
kN
)⋂(· · · (B+
knB−
kn+1∪ B−
knB+
kn+1) · · ·
),
which are equivalent to(Bs1
k1. . . Bsn
knBsn
kn+1∪ B−s1
k1. . . B−sn
knB−sn
kn+1
)B±
kn+2. . . B±
kN,
and (Bs1
k1. . . Bsn
knB−sn
kn+1∪ B−s1
k1. . . B−sn
knBsn
kn+1
)B±
kn+2. . . B±
kN,
respectively. Therefore, the 2n subbands of the n-level DEW can be written as⋃si∈{+,−}
(Bs1
k1. . . B
sn+1
kn+1∪ B−s1
k1. . . B
−sn+1
kn+1
)B±
kn+2. . . B±
kN,
April 1, 2005 DRAFT
26
which proves the claim in the theorem for all n ∈ {1, . . . , N − 1}.
REFERENCES
[1] I. Daubechies, Ten Lectures on Wavelets. Philadelphia, PA: SIAM, 1992, notes from the 1990 CBMS-NSF Conference
on Wavelets and Applications at Lowell, MA.
[2] M. Vetterli and J. Kovacevic, Wavelets and Subband Coding. Prentice-Hall, 1995.
[3] S. Mallat, A Wavelet Tour of Signal Processing. Academic Press, 1998.
[4] J. Daugman, “Two-dimensional spectral analysis of cortical receptive field profile,” Vision Research, vol. 20, pp. 847–856,
1980.
[5] E. P. Simoncelli, W. T. Freeman, E. H. Adelson, and D. J. Heeger, “Shiftable multiscale transforms,” IEEE Trans. Inform.
Th., Special Issue on Wavelet Transforms and Multiresolution Signal Analysis, vol. 38, no. 2, pp. 587–607, March 1992.
[6] R. H. Bamberger and M. J. T. Smith, “A filter bank for the directional decomposition of images: theory and design,” IEEE
Trans. Signal Proc., vol. 40, no. 4, pp. 882–893, April 1992.
[7] J. P. Antoine, P. Carrette, R. Murenzi, and B. Piette, “Image analysis with two-dimensional continuous wavelet transform,”
Signal Processing, vol. 31, pp. 241–272, 1993.
[8] F. G. Meyer and R. R. Coifman, “Brushlets: a tool for directional image analysis and image compression,” Journal of
Appl. and Comput. Harmonic Analysis, vol. 5, pp. 147–187, 1997.
[9] N. Kingsbury, “Complex wavelets for shift invariant analysis and filtering of signals,” Journal of Appl. and Comput.
Harmonic Analysis, vol. 10, pp. 234–253, 2001.
[10] I. W. Selesnick, “The double-density dual-tree DWT,” IEEE Trans. Signal Proc., vol. 52, no. 5, pp. 1304–1314, May 2004.
[11] R. G. Baraniuk, N. Kingsbury, and I. W. Selesnick, “The dual-tree complex wavelet transform - a coherent framework for
multiscale signal and image processing,” IEEE SP Mag., submitted, 2004.
[12] E. J. Candes and D. L. Donoho, “Curvelets – a suprisingly effective nonadaptive representation for objects with edges,”
in Curve and Surface Fitting, A. Cohen, C. Rabut, and L. L. Schumaker, Eds. Saint-Malo: Vanderbilt University Press,
1999.
[13] M. N. Do and M. Vetterli, “The contourlet transform: an efficient directional multiresolution image representation,” IEEE
Trans. Image Proc., to appear, http://www.ifp.uiuc.edu/˜minhdo/publications.
[14] Y. Lu and M. N. Do, “CRISP-Contourlets: a critically-sampled directional multiresultion image representation,” in Proc.
of SPIE conference on Wavelet Applications in Signal and Image Processing X, San Diego, USA, 2003.
[15] ——, “The finer directional wavelet transform,” in Proc. IEEE Int. Conf. Acoust., Speech, and Signal Proc., Philadelphia,
2005.
[16] P. P. Vaidyanathan, Multirate Systems and Filter Banks. Prentice Hall, 1993.
[17] M. Holschneider, R. Kronland-Martinet, J. Morlet, and P. Tchamitchian, “A real-time algorithm for signal analysis with the
help of the wavelet transform,” in Wavelets, Time-Frequency Methods and Phase Space. Springer-Verlag, Berlin, 1989,
pp. 289–297.
[18] F. Fernandes, R. van Spaendonck, and C. Burrus, “A new framework for complex wavelet transforms,” IEEE Trans. Signal
Proc., vol. 51, no. 7, pp. 1825 – 1837, July 2003.
DRAFT April 1, 2005
LU AND DO: A DIRECTIONAL EXTENSION FOR MULTIDIMENSIONAL WAVELET TRANSFORMS 27
[19] ——, “Multidimensional, mapping-based complex wavelet transforms,” IEEE Trans. Image Proc., vol. 14, no. 1, pp. 110
– 124, January 2005.
[20] W. Chan, H. Choi, and R. Baraniuk, “Directional hypercomplex wavelets for multidimensional signal analysis and
processing,” in Proc. IEEE Int. Conf. Acoust., Speech, and Signal Proc., Montreal, Canada, May 2004.
[21] S.-M. Phoong, C. W. Kim, P. P. Vaidyanathan, and R. Ansari, “A new class of two-channel biorthogonal filter banks and
wavelet bases,” IEEE Trans. Signal Proc., vol. 43, no. 3, pp. 649–665, Mar. 1995.
[22] T. Laakso, V. Valimaki, M. Karjalainen, and U. Laine, “Splitting the unit delay - tools for fractional delay filter design,”
IEEE SP Mag., vol. 13, no. 1, pp. 30 – 60, January 1996.
[23] A. V. Oppenheim, R. W. Schafer, and J. R. Buck, Discrete-Time Signal Processing, 2nd ed. Prentice-Hall, 1999.
[24] D. Donoho and I. Johnstone, “Ideal spatial adaptation via wavelet shrinkage,” Biometrika, pp. 425–455, December 1994.
April 1, 2005 DRAFT