The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
1262
multi-spectrum image and full-color image fusion system are as
shown in Figure 1.
3. WEIGHTING IMAGE FUSION BASED ON
WAVELET TRANSFORMATION
The fusion method based on the wavelet transformation has
retained and inherited the main merits of tower-shaped
transformation fusion method [2-4]. Simultaneously, because
the orthogonal wavelet transformation has non-redundancy, the
image data quantity after wavelet transformation will not
increase. By using the direction of wavelet transformation, a
better visual effect of fusion image will be able to obtain based
on the vision characteristics that the human eye have different
resolutions to high frequency component of different directions.
3.1 Image fusion based on wavelet transformation
The image fusion realization requests to extract information
(details) from each source image and obtains the effective
demonstration in the final fusion image. According to the
general theory of imagery processing, the image information
(details) is concluded in the image high frequency component,
therefore, the key point of image fusion research is to seek an
Image A
Raw imagery after Multi-dimensional
matching resolution
Figure2 Procedure of wavelet multi-dimensional fusion
(1) Carry on the wavelet transformation separately to each
source image to establish various images of the wavelet tower
shaped transformation.
(2) Carry on fusion processing separately to each
transformation level and to different frequency components of
each transformation level using different fusion operator, to
finally obtain fusion wavelet pyramid.
(3) Carry on the wavelet inverse transformation (i.e. to carry on
image restructuring) to wavelet pyramid after fusion to obtain
restructuring image namely for fusion image.
When fusing the multi-spectrum image and the high spatial
resolution image, David proposed one method (which was
afterwards called the WT method), which simultaneously
carries on the wavelet transformation to the multi-spectrum
image and the high spatial resolution image, combines the low
frequency component of multi-spectrum image with the high
frequency component of high spatial resolution image, then
carries on the wavelet inversion, and finally obtains a multi
spectrum image of the high spatial resolution. That is the initial
wavelet transformation fusion law [5].
appropriate processing method of fusing detailed information of
source image respectively, that is, how will the information be
fuse-processed effectively in the corresponding frequency band.
According to the multi-resolution analysis theory, the source
image after the wavelet transformation, the inner tube signal
row Z)j j ,Dj.j , Dj j, contains separately the image high
frequency component of corresponding frequency directed in
normal, horizontal and 45° of the image, therefore, using the
image tower-shaped structure after wavelets decomposing, to
carry on fusion processing separately according to different
transformation level and different frequency band, the
information detail from different images can be fused
effectively.
Carry on N wavelet transformation to the two-dimensional
picture, and it will finally get some (3N+1) different frequency
bands, which contains a 3N high frequency bands and a low-
frequency band. The image fusion program based on wavelet
multi-dimensional transformation is shown as in Figure 2. Take
two image fusions as an example, many image fusion methods
may be analogized from this regarding. Suppose A and B are
two primitive images, F is the image after fusion, its basic
fusion steps are as follows:
Image F
Wavelet
►
inverse
transforrr
Imagery
after fusion
In practical application, in order to prevent the spectrum
distortion caused by replacement in the fusion process, it is
necessary to enlarge various relevance of wave bands image
before the transformation wavelet through the histogram
matched various wave band images, to reduce the spectrum
deviation of fusion image.
The WT method is always considered as the superior high
resolution and the multi-spectrum image fusion method. But it
has also a shortcoming: Because the wavelet transformation has
localization in the good air zone and the frequency range, it
may well retain the spectrum information in the multi-spectrum
image, but because this method has discarded low frequency
component of the high spatial resolution image, it results in the
ringing effect in the inverse transformation.
3.2 Weighting image fusion algorithm based on wavelet
transformation
As we know, the spatial variation surface of the remote sensing
image grey level expresses has the characteristics of stochastic
changes. But analyzed from the spatial frequency spectrum
angle, they may all be transformed to the spectrum which is
composed of different spatial frequency spectrum waves.