The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
forward and inverse transformation and just concentrates on the 
calculation ( 
requirements. 
calculation of 60kx and ^ iJ) , decreasing computational 
The experimental data consists of a slice of IKONOS 
panchromatic band with 1761 x 1649 pixels, and the 
cooresponding multispectral image with 441 x 413 pixels 
including B, G, R, NIR bands. We write Matlab programs to 
perform the fusion operations using the regular method and 
generalized model based method, respectively, in the same 
hardware and software platform. The experimental results show 
that the runtime not including resampling time of the 
multispectral image, input and output time for the new 
implementation is 1.0938s, while the time for the regular 
implementation is 1.3906s, saving 0.2969s or 21.35%. At the 
same time, the pixel value of the two types of fusion results, 
shown in fig. 1 (c) for regular implementation and fig. 1 (d) for 
the generalized model based implementation, is the same. Fig.l 
(a) and fig.l (b) are the original multispectral image and PCA 
fusion results, respectively. 
CONCLUSIONS 
This paper presents the generalized model for remotely sensed 
data pixel-level fusion, which can clearly describe the 
relationships among the original multispectral image, the spatial 
details extracted from the high-resolution panchromatic image,