The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
1263 
Regarding the different remote sensing images in the identical 
area, its difference is not in the low frequency part, but in high- 
frequency part. In other words, its low frequency part of the 
spatial frequency spectrum is the same or similar, and the 
remarkable difference is only in the high-frequency part. 
Therefore, it can be processed according to the low frequency 
and the high frequency by using different fusion rules. 
To the low frequency components, T LL P JL , we use the 
simple weighted average operator, namely: 
Tll ~ Kt xT ll +K p xP ll , in which K T K p are 
The comparison fusion process is as follows: 
(1) Determine a size of an air zone window (for instance, may 
take 5x5), and then divide W p and W T into some sub 
block images of the size of this window. 
(2) Carry on the value distribution statistics to each sub-block 
image, and calculate its mean value and the standard deviation. 
(3) Determine the data fusion value of each sub-block image as 
W'= fJ. p X W p + jU T + W T , in which /Up and jUj is 
the weighting factor of each sub-block of W p and W T . If 
the fruit block standard deviation of W p is bigger than the 
image sub-block standard deviation of W Tj then 
ju p > ju T , else Pp(ju T . 
Recalculate (2) and (3), to get the fusion value of all sub-block 
images. Thus the new high-frequency unit W' is obtained. Then 
add the high-frequency component and the low frequency 
ingredient to obtain the fusion low frequency component, 
weight coefficient. Because the low frequency part is mainly 
taken from multi-spectrum image, K T )K p . And To the high 
frequency components, we proposed a method according to the 
image statistical property to process the fusion based on the 
biggest criterion to partial region standard deviation [6]. 
Divide two high frequency images into certain sub-block 
images, and then compare the standard deviation of 
corresponding sub-block images. The bigger the standard 
deviation of a sub-block image, the bigger the weight is. 
namely: T L — T LL + W’ . Finally combine this new low 
frequency component with the high frequency component of 
high spatial resolution image to achieve the final fusion result in 
wavelet inversion. 
Avoid footnotes, but if you need them, mark footnotes in the 
text with an asterisk (*); use a double asterisk (**) for a second 
footnote on the same page. Place footnotes at the bottom of the 
page, separated from the text above it by a horizontal line. 
3.3 Fusion result appraisal 
In order to confirm the accuracy and the validity of the method, 
we have selected some region terrain general picture Landsat 
TM multi-spectrum image and the SPOT image to carry on the 
fusion experiment. First we carry on the strict matching and 
processing, analyze respectively the wavelet of the two images, 
and then process it according to the method as above-mentioned. 
Thus the final fusion image is obtained. In order to be compared, 
these two images are processed with the WT method. The result 
is shown in figure 3. 
(1) Multi-spectral image 
(2) high-resolution image 
The results of different fusion method may be judged visually. 
The merit is direct, simply, which makes the qualitative 
appraisal direct according to the contrast after imagery 
processing: The shortcoming is its strong subjectivity. In order 
to evaluate the fusion effect objectively and quantitatively, we 
analyzed the information contained in the fusing image. We use 
the visual distinction firstly, and then we carry on the 
quantitative analysis. (1) is the Landsat TM multi-spectrum