P = AxL + BxL +CxL +DxL 
Blue Green Red N 
(3) 
Where P = approximated panchromatic 
L =Top-of-atmosphere band-integrated radiance 
of blue band 
L = top-of-atmosphere band-integrated radiance 
of green band 
L r ; =top-of-atmosphere band-integrated radiance 
of red band 
L =top-of-atmosphere band-integrated radiance 
of near-infrared band 
A,B,C,D = Weighted coefficients, which can be 
evaluated according to the image scene context by 
iterative computing or giving a estimated constant. 
The new pan-sharpening algorithm can be shown as follows 
(Figure 3). 
4. APPLICATION AND DISCUSSION 
As an example, the new pan-sharpening algorithm and other 
data fusion methods like HIS, PCA and UNB (Digital Globe’s 
default pan sharpening algorithm) were applied to four related 
multi-spectral images of a Quickbird satellite scene to extract 
geographic features information (see Figure 4). 
The PCA algorithm’s sharpness is too bad in scenes where 
there is a lot of green vegetation; other three pan-sharpening 
algorithms have a good sharpness in the vegetation. The road is 
very clearly in IHS, UNB and new pan-sharpening algorithms 
and is some blurry in PCA algorithm. 
The HIS algorithm’s color recovery is worse than UNB and 
new pan-sharpening algorithms in whole scene. The new 
pan-sharpening algorithm’s color recovery is good in whole 
scene. 
Some tiny objects like car etc al are more clearly and easily to 
observe and analyze in new pan-sharpening algorithm than in 
PCA, UNB, HIS algorithms. 
Figure 4. Comparison of HIS, PCA, UNB and new 
pan-sharpening algorithms (from upper left to lower right) 
In theory, the P/P’ (see Figure 3) should equal or very nearly to 
1. A,B,C,D can work better in a scene with single 
geographic features , we can evaluate the weighted coefficients 
a constant according to the geographic features types. The 
more convergence of P/P’, the better is the pan-sharpening 
result (see Figure 5). 
When the scene is full of various geographic features, the 
weighted coefficients cannot evaluate a constant, the value of 
weighted coefficients change with respect to the pixel of 
geographic features. 
Figure 5. Comparison of new pan-sharpening algorithms with 
different weighted coefficients (Left: A = 0.4,B = 
0.63,01.6,D=1.45; Right: A = 0.4,B = 0.6,C=1.5,D=2.0) 
236 
5. CONCLUSION