The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
(d) triangle IHS fused image 
4. EXPERIMENT AND RESULTS ANALYSIS 
For the sake of validating the effectiveness of the algorithm, we 
employed a RADARSAT fine model image, with 10m* 10m 
resolution, as the original SAR image, and a fused SPOT-5 
image, with 2.5m*2.5m resolution, as the original multispectral 
image. The fused SPOT-5 image is synthesized by the SPOT-5 
multispectral bands with 10m* 10m resolution and the SPOT-5 
panchromatic band with 2.5m*2.5m resolution. As we all 
known, the image fusion usually occurs with a high resolution 
panchromatic image and a low resolution multispectral image, 
yet it is opposite in this paper. Both of these strategies are based 
on the fusion destination, in the usual situation, we want to 
acquire an image holding high spatial resolution and fine 
spectral information; while in the application-oriented situation 
we assumed, we would like to gain an image with sufficient 
spectral information and as more SAR information as possible. 
So the original images are just designed for this purpose. 
The comparison is done between the novel fusion method and 
other traditional fusion methods, such as Brovey (Zhang,2006), 
triangle IHS (Wang,2000), PCA (Jia,1998), Gram-Schmidt 
(Li,2004) and wavelet transform method in article (Li, 1994). 
With respect to the fusion method based on wavelet transform 
in this paper, both the normal wavelet transform method and the 
novel method adopted the sym4 wavelet base. The window size 
in the novel method is 7*7 pixels. The image size is 1024*1024 
pixels. 
The original and result images are shown in Figure.2, where, 
Figure2(a) is RADARSAT image, Figure2(b) is multispectral 
image, Figure2(c)-(h) are the fused image of Brovey, triangle 
IHS, PCA , Gram-Schmidt, normal wavelet transform, and the 
novel method, respectively. 
are done for SAR image S, multispectral image X and 
normalized weight image , and pyramid image 
S pym , x pym and w/ ym , respectively, are obtained. 
6. In wavelet domain, pyramid images S P '"' and are 
fused according to the fprmula below, and fused pyramid 
image F py "' in wavelet domain is obtained. 
r, S""(2’ ,x,y) + W i n "(2 I ,x,y)xX? m (2 i ,x,y) 
F t (2 \x,y) = * i — 
l + Wr(2 J ,x,y) (5) 
where,y=l,2,...,Jis pyramid level, 2 1 is resolution, and 
F” m (2', x, y) is the value of pyramid image ony-level at 
(x, y) position. 
7. By reconstructing pyramid image F p>m , fused image of 
the Mi band F k is ultimately obtained. Hence, the whole 
fusion image is gained. 
As we can see from the algorithm, along with moving of the 
window, the fusion image might auto-adaptively fuse 
information of the two images according to the information 
capacities of SAR image and multispectral image, more 
sufficiently considering local features of images, but not 
perform the whole image with a single and ponderous way. 
(c) Brovey fused image