The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
1264
image, (2) is the SPOT image, (3), (4) is the fusion image in
WT method when wavelet transformation exponent number J=l,
2, (5) fusion image which obtains for this article method.
From the chart we may see, the result basically maintained the
original map spectral characteristic regarding WT method when
J=l, but image spatial detail performance ability is not high, the
entire image is quite dim, the edge characteristic is not very
clear like rivers edges, the enhancement effect is not very good.
Regarding the result of WT method J=2 fusion, the ringing
effect appeared, to create the fuzzy texture. By the method
proposed in this paper we obtained the image (5) in which not
only the outline is quite clear but the river boundaries is clearer
than (3) and (4). Moreover the spectral characteristic maintains
quite well.
When we carry on the quota analysis comparison, we use the
correlation coefficient and spectrum tortuosity of the image to
carry on the appraisal. Table 1 gives the parameter contrast of
the image fusion results, and it can be easy to see from the
parameters in the table that the method proposed in this paper
has a very good effect in enhancing the spatial texture
characteristic of multi-spectrum images and in maintaining
spectrum information.
Method
types
spectrum
tortuosity
Correlation coefficient
SPOT
Image
TM multi-
spectral
Image
Method
J=1
WT
19.03
0.61
0.79
Method
J=2
WT
24.42
0.72
0.74
The
method
new
24.05
0.83
0.74
Table 1 Comparison of images fusion result
4. MULTI-SPECTRUM IMAGE FUSION BASED ON
WAVELET TRANSFORMATION AND VISUAL
CHARACTERISTICS
Although the fusion effect of wavelet transformation method is
relatively ideal, its transformation restructuring is actually a
process of the high pass and the low pass filter, and, to a certain
extent, loses some edge information in the primitive image, as a
result of which the ringing effect in fusion image appears.
Therefore, when the image fusion is carried on, the wavelet
transformation and other alternative means are often combined.
This paper proposes a new image fusion method in which the
wavelet transformation is combined with the features of human
vision system. This method can not only enhance the spatial
detail ability of multi-spectrum image, maintenance the
spectrum information, but also avoid strengthening the result
ringing effect.
are fully used in the imagery processing, the image quality will
be greatly improved [7-8].
4.2 Fusion algorithm based on wavelet transformation and
features of human vision system
Fusion method theory is based on: Through the statistics of the
value distribution, the sub-graph of source image A and B after
wavelet transformation has such characteristics:
(1 )The data change scope in the original map region is identical
to that of the sub-graph corresponding region.
(2) Regarding the different source images of identical goal or
object, the data value of its low frequency image in
corresponding region is the same or similar, but there is a
remarkable difference in the high frequency sub-graph. The
above-mentioned characteristic of the wavelet transformation
has provided the theory basis for the choice of the effective
fusion method. As for combining the wavelet transformation
with the feature of human vision system, a new fusion method
is brought up in this paper.
(1) Carry on the geometry correction to the multi-spectrum
image and the high resolution image respectively, use the
geometry matching method based on the region, and then match
the multi-spectrum image with the high resolution image.
(2) Carry on a step wavelet resolving to the matched multi
spectrum image and the high resolution image to obtain
respectively the low frequency component and the high
frequency detail component.
(3) Carry on the uniformity method fusion to the two
corresponding images of the high frequency and the low
frequency sub-graph respectively, the processes are:
(D Dissect the two images into some sizes for NxN blocks.
Suppose Aj and Bj expresses the block of image A and B
separately.
(2) Calculate the uniformity measure of each block according to
the formula which literature [9-10] provides. Suppose J(Ai) and
J(Bi) are respectively the uniformity measures of Ai and Bi.
(3) Compare the uniformity measure of the corresponding block
of the two images, and obtain the fusion image ith block Fi
F ; =
4
B,
(A,+B,)/2
J(A, )> J(B t ) + TH ^
J(A t )< J{B t )-TH
otherwise
Among them, TH is the threshold value parameter.
4.1 Human eye vision system
The image is to be looked at by humans, therefore the imagery
processing should follow the features of human vision system.
The vision is the result that images are left in human eyes. The
complexity of visual processing has not have been understood
and mastered by human beings at present. But people have
discovered some visual phenomena, such as visual threshold,
visual masking effect. If these features of human vision system
(D In turn carry on the above operation to all image blocks, the
new fusion image is obtained.
(5) Carry on the wavelet inversion to the obtained high
frequency and the low frequency sub-graph, and obtain the
fused multi-spectrum image.