220 
In the research of Munechika et al. (1993), simulated 
synthesized panchromatic images are calculated by method 
proposed by Suits et al. (1988): 
Pan Lsyn=Y. ( Pi XS Li (2) 
i=l 
To obtain parameter^?., Munechika et al. (1993) adopted an 
improved atmospheric model to modify received object 
reflection and relative spectral response, and to measure 5 
representative reflection spectrum of land cover, including city, 
soil, water, tree and grass, cp, can be obtained through 
regression analysis calculation of 5 types actual measure value 
and value from atmospheric model. After get Pan Uvn , 
panchromatic images are processed histogram matching with 
simulated synthesized images to eliminate difference between 
atmospheric and radiation. 
Pan H . panchromatic image, not only with band 1, 2, 3, 4 of 
TM image, which can avoid TM image colour distortion. 
2.3 SSVR Method 
There still exist some problems of SVR improved by Zhang 
(1999) and Zhang (2001). Firstly, according to the improvement, 
the less correlation lies, the more spectral information can be 
reserved, and the accuracy of regression coefficient which is 
calculated through regression analysis method is doubtful. 
Secondly, numerous pixels with different types of land cover 
which is selected randomly should be added to regression 
analysis, so that more experiences are needed. Thirdly, 
parameters obtained from regression analysis have no physical 
meaning and cannot be explained, on the same time, the 
Simplified Synthetic Variable Ratio (SSVR) is presented. 
2.2 SVR Method improved by Zhang 
Zhang (1999) and Zhang (2001) suggest that it is difficult to get 
parameter (p. in SVR method of Munechika et al.( 1993), and 
only several specific land cover types are used. For one thing, 
this method is not practical in regular large area fusion. For 
another thing, 5 types land cover cannot be representative in a 
large area, and cannot assure that spectra properties of other 
types land cover can be recorded. Furthermore, SPOT 
panchromatic images should be processed histogram matching, 
as Pan Lsvn in equation (2) can be obtained through certain 
atmosphere and specific land cover type. 
In view of above problems, Zhang (1999) and Zhang (2001) 
modified SVR method in this way: 
XSP, = Pan 
XS, 
(3) 
Pan 
Hsyn 
here, XSP i and Pan H remain the same meaning as in equation 
(!)• XS H means band i grey value after resampling 
multispectral images to the same pixel size with panchromatic 
images. p a n Hsyn means grey value of high spatial resolution 
image which is synthesized in the way shows in below equation: 
Pan Hsyn ='Z ( P‘ XS h, ( 4 ) 
Equation (5) is used to calculate parameter q>. which can be 
obtained through regression analysis calculation of 5 original 
PAN images and original MS images XS Hi (has the same pixel 
size with Pan H )• 
Pan h =Yu ( Pi XS m ( 5 ) 
The core problems of SVR and its modified method are 
simulated from low spatial resolution multiple bands to low 
spatial resolution panchromatic bands. The SSVR method doing 
the simplification is based on some rules: assuming Pan images 
have the same spectrum range with MS images, which is one of 
the hypothesis of ISVR, the Pan energy simulated through low 
resolution MS images can be approximately substituted by the 
energy of the sampled high resolution Pan images, that 
is e « E 
^ Pan ^ ~ ^ Pan L 
The equation of SVR could be expressed as 
XSP, = Pan, 
V P Pan 
(6) 
where XSP, is radiance value of band i of the high resolution 
images after fusion, Pan H is the radiance value of original 
high resolution Pan images, E i s the i band radiance value 
of the original low resolution multispectral images, E Pan is 
radiance value resampled from high resolution Pan images, and 
represents the energy ratio should be converted from 
E 
V Pan i 2 
low resolution to high resolution on account of the 
correspondence among pixels between high resolution images 
and low resolution images. Pan H denotes the spatial 
information of fused images, and spectral information is 
embodied in the form of 
'N 
"i 2 
Mean value method is adopted resampled from E to E 
P an h p an L 
Y E 
L-t pan h 
(7) 
N 
Improvements of SVR method by Zhang include: more stable 
parameter (p. , easier acquisition of parameter (p., and the merit 
that in fusion process, every band can be synthesized with 
where N denotes the pixel amount of low resolution 
corresponding to high resolution. If Pan L and E pan are 
corresponded to Figure 2-a and Figure 2-b respectively, then