208
only the higher resolution image is decomposed or analyzed.
Then the low frequency image of the higher resolution image is
replaced by low frequency image of the lower resolution image
or the lower resolution image. Finally, an inverse wavelet
transform is applied. The result is an image which merges the
higher resolution image and the lower resolution image.
In this paper, the conventional wavelet-based fusion method is
modified. Before wavelet transforms, the multi-spectral image
is transformed to gray scale image. There are two methods: IHS
and PC. The IHS method accepts only 3 input bands. It has
been suggested that this technique produces an output image
that is the best for visual interpretation. But Yocky (Yocky,
1995) demonstrates that the IHS transform can distort colors,
particularly red, and discusses theoretical explanations. But the
PC Method will accept any number of input data layers. It has
been suggested that this technique produces an output image
that better preserves the spectral integrity of the input dataset.
Thus, this method would be most appropriate if further
processing of the data is intended; for example, if the next step
was a classification operation. In this paper, the purpose of
fusion is to further apply it, so the PC method integrated with
wavelet is used to fusion the SAR image and TM. Thus the
fused image can not only preserve the spectral information
better,but also reduce the information redundancy.
2.3 Fusion Image Quality Appraisal
In this paper several parameters including mean, entropy and
standard deviation are adopted to value quality of the fusion
image. Mean, standard deviation and correlation are parameter
we are familiar with, so entropy is only presented in this paper.
According to Shannon information theory , the larger the
entropy of image is , the richer the information and the better
quality of a image is .
H(x)=-I.P : iog 2 P l (7)
/=0
where H(x) = the entropy of image
i = the grey value of pixel
n = the number of pixel of image
Pj =the probability of i
3. EXAMPLE AND ANALYSIS OF FUSION IMAGE
In the following sets of example, wavelet-based fusion is
applied to 10m resolution SAR image and 30m resolution
Landsat multi-spectral TM image (4, 3,2band). The process is
as follows (Figure2).The fusion products of different methods
are demonstrated (Figure4).In comparison to the original
images (TM and SAR), the fused image includes spectral
information and detail information. In other words, the new
image contains the spatial detail feature of high resolution SAR
image and spectral information of multi-spectral TM image.
And the fused image based on wavelet integrated PC has a
better visual effect compared with result of other fused method
In addition to the visual analysis, we extended our investigation
to a quantitative analysis. In order to appraise the fused image
quality, we adopt several parameters to analyze the fused
images, including mean, standard deviation, correlation and
entropy which were used in other studies (Wald et al, 1998).
Figure 2 Schematic of Wavelet integrated PC
Table 1 the comparison of fused image of different methods
Standard
Method
Mean
deviation
correlation
Entropy
98.468
78.861
0.923
116.835
88.744
0.258
PCA
109.659
55.457
0.035
7.839
97.752
74.408
0.726
111.679
83.315
0.812
IHS
154.965
87.669
0.915
7.704
79.036
79.808
0.873
Brovey
72.186
73.094
0.185
7.344
81.600
52.760
0.100
Wavelet
104.075
88.847
0.674
integrated
110.925
82.048
0.146
7.847
PC
105.985
58.115
0.033
Original
79.794
77.443
0.940
Image
103.602
88.555
0.465
7.5528
(TM)
105.593
56.121
0.268
Table 1 presents a comparison of the result of image fusion
using mean, standard deviation, correlation and entropy. In
general the standard deviation can reflect the information of
image and deviation from original image to a certain degree. If
from the view of standard deviation only, the fused image based
on IHS is the most, but the correlation coefficient is also the