ul 2004
'ssential
the step
t, while
8a) for a
does not
vavelets.
de appr-
PR and
y compl-
it from a
At level
) tree B.
they are
nce the
el, whic-
nalysis.
, and the
ymmetry
ollowing
Tree B
x wavelet
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
3. THE 2-D DUAL-TREE COMPLEX WAVELET
TRANSFORM
For 2-D signals, we can filter separately along columns and then
rows by the way like 1-D. Kingsbury figured out in (Nick
Kingsbury: 11998a) that, to represent fully a real 2-D signal, we
must filter with complex conjugates of the column and row
filters. So it gives 4:1 redundancy in the transform. Furthermore,
it remains computationally efficient, since actually it is close to
a classical real 2-D wavelet transform at each scale in one tree,
and the discrete transform can be implemented by a ladder filter
structure.
The quad-tree transform is designed to be, as much as possible,
translation invariant. It means that if we decide to keep only the
details or the approximation of a given scale, removing all other
scales, shifting the input image only produces a shift of the
reconstructed filtered image, without aliasing. ( A. Jalobeanu ,
2000)
The most important property of CWT is that it can separate
more directions than the real wavelet transform. The 2-D DWT
produces three bandpass subimages at each level, which are
corresponding to LH, HH, HL, and oriented at angles of 0? , +
45°, 90°. The 2-D CWT can provide six subimages in two
adjacent spectral quadrants at each level, which are oriented at
angles of +15°, +45°, +75°. This is shown in fig 3. The strong
orientation occurs because the complex filters are asymmetry
responses. They can separate positive frequencies from negative
ones vertically and horizontally. So positive and negative
frequencies won't be aliasing. The orientations of details is
shown in fig 4. Fig 5 shows the transform of an isotropic
synthetic image at level 3, which also contains details at
different scales. The orientation selectivity is more clear under
each scale in comparison with the classical wavelet transform.
Since CWT has so many advantages, we consider use CWT to
carry out image fusion instead of DWT. Then we design an
image fusion method based on CWT in next section.
4. ANIMAGE FUSION APPRAOCH BASED ON CWT
The DWT has already been used for image fusion ten years ago.
Though image fusion approaches by wavelet transform have
been improved to be adaptive to process varied images, two
disadvantages (lack of shift invariance and poor direction
selectivity) still exist. They have hampered the further
application of wavelet transform in image fusion.
The CWT is a good solution to this problem. It is approximate
shift invariant. If the input signal shift a few samples, the fused
image will be reconstructed without aliasing, which is useful to
the not strictly registered images. Morover it can separate
positive and negative frequencies and provide 6 subimages
with different directions at each scale. So the details of CWT
can conserve more spatial information than DWT. The spatial
531
REAL DW
Figure 3. 2-D impulse responses of the complex wavelets at
level 4 (6 bands at angles from -75° to +75°) and equivalent
responses for a real wavelet transform (3 bands)
-N2
X
IN.
/
Figure 4. Directional selectivity of the frequency space
corresponding to the complex wavelet transform
Figure 5. Left : isotropic test image containing various scale
information, right: magnitude ofits complex wavelet transform
at level 3 showing both directional and scaling properties
can conserve more spatial information than DWT. The spatial
resolution of the fused image is more closer to the
high-resolution image. PR, limited redundancy and high
computation efficiency make it suitable for image fusion.
We design an approach based on the quad-tree complex wavelet
transform for fusing a low resolution multi-spectral image and a
high resolution panchromatic image. First the registered
multi-spectral image and panchromatic image are decomposed
by complex wavelet respectively , then the approximate and
detail parts of two images are fused according to some rule at
each level, finally the fused image is reconstructed. This is
illustrated by fig 6. The fusion procedure can be described
detailedly as following:
(1)Each band of the low resolution multi-spectral image and the
high resolution panchromatic image are geometrically
registered to each other. After geometrical rectification , ihe
images have the same size.
(2)The panchromatic image is stretched tally with each band
of multi-spectral images respectively according to the
histogram.
(3)Decomposed the histogram-specified panchromatic image