International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
  
  
  
  
| 3x(i-1) 3x(n-i)-16 4444244448 | 
So Sor Ag Aor Sor A 7 
A4j43 =| 0.0,A ,0,- Le A m „00.2.0- ZZ EL 
clon; Olatj Ch; acy aco p oco, | 
| 
| | 
644444474444448 6447448 
X =| Qlong, lato, hg, ^ , lon, , lat, 1,009, óc, Óco,. 
6444449444448 647248 
Xo 7 | long,lato, ho, ^ lon, 4, lat, 4,09 ,, 00 p. Co, (9) 
Yo aX tX (10) 
new 
In equation 9, there are no derivation terms to óh, ,, which is 
the height of ground point. 
Since the inverse f; g;, f» and g» are set of functions, some of 
which are iterative functions, their partial derivatives in A 
cannot be directly derived. They are instead analytically 
computed. 
With the initial A; of tie points to 0, /on;,lat; of both tie points 
and ground point calculated from equation 2, and 3 coefficients 
to 0, X can be solved with equation 8. Then the initial values are 
replaced with the new values X,,,, with equation 10. The least 
square estimation is repeated until Æ is stabilised to a minimal. 
4. IMAGE ORTHORECTIFICATION 
The original sensor model of PAN and the refined sensor model 
of XS images are used to orthorectified PAN and XS images 
respectively. As mentioned earlier, the look angles between the 
PAN and XS Image for SPOT 5 HRG instrument is 1.06?. This 
is equivalent to a B/H ratio of less than 0.02. If the error in the 
DEM used is 100m in height, the relief displacement error 
between the PAN and XS images will be about 2m, less than a 
PAN pixel size. This insensitivity to DEM accuracy suggests 
that a coarse orthorectification with the km gridded GLOBE or 
SRTM will be accurate enough for removing the relative 
displacement error to sub pixel accuracies. 
5. COLOR MERGING 
Merging of the high resolution colour imagery is performed by 
multiplying each spectral band with a sharpening factor 
computed from the intensity values of the panchromatic image 
and the corresponding pixels of the multispectral images. The 
equation for merging is described below: 
  
Lan 
la m 1 (11) 
merge I lys 
XS 
where i =the band index of SPOTS imagery 
Li and IS = intensity of panchromatic and 
multispectral imagery, respectively 
Ly and I, = grey value of each band in 
merge = 
multispectral and merged imagery, respectively 
The factor ( / 
pon td fc) 18 referred to as the sharpening factor. 
Since the spectral range of the PAN image overlaps only the 
RED and GREEN bands of the XS image, we compute the 
T by taking averaging of pixel values of only the RED and 
XN 
GREEN bands. The non-overlapping bands (NIR band and 
SWIR band) can either be sharpened with the same sharpening 
factor or merely resampled to the resolution of the 
panchromatic imagery. The pseudo-natural colour SPOT 5 
imagery can also be merged with the panchromatic imagery in 
the same way. 
6. TEST IMAGERY AND RESULTS 
To test the method developed, we chose a site in Bandung, 
Indonesia, where there are sullen volcanoes and the terrain 
height variation from a few tens to more than 3000 meters. A 
pair of 2.5m PAN and 10m XS SPOT 5 images taken from the 
same instrument were selected over the area. Figure 2 and 
Figure 4 show the overview of level 1A images of PAN and XS 
respectively. Parts of the full resolution images are shown in 
Figure 3 and 5. 
The edges of PAN imagery detected by Canny detector are 
shown in Figure 6 in grey, while the corners detected by Harris 
corner detector are shown in black in the same figure. Total 40 
image patches have been matched and the tie points selected 
automatically for the sensor model refinement. The rms 
residual after least squares solution of the refined XS sensor 
model is 0.7 pixel (PAN image pixel size). 
Figure 7 shows the overview of orthorectified PAN image after 
applying its sensor model. And Figure 8 shows the overview of 
orthorectified XS image after applying its refined sensor model. 
The final merged 2.5m pixel pseudo-natural color imagery 
(merged from both orthorectified PAN and XS images) is show 
in Figure 9 in overview and Figure 10 in full resolution. 
7. CONCLUSION 
image registration method 
and corners) and 
To conclude, an automated relative 
based on feature information (edges 
refinement of XS sensor model relative to the PAN sensor 
model is described. The GLOBE or SRTM DEM used in 
orthorectification was sufficient to eliminate the relative relief 
displacement to better than one PAN pixel. The PAN and XS 
orthorectified images generated can then be merged to produce 
a high resolution multispectral image, that has the high 
resolution of PAN and colour information of XS. The method 
has been tested on a few SPOTS PAN and XS image pairs of 
various incidence angles and various relief situations. The 
results were consistent. 
We expect the automatic matching method to have problems in 
imagery where there are too much clouds or water. In such 
situations, the program may not be able to find enough image 
patches with good edges and corners content. An option has 
been provided to manually provide the corresponding image 
Intern. 
MIEL IE 
registi 
locate 
refinit 
way. 
Refer 
Bergh 
patter 
pp. 72 
Canny 
IEEE 
PAMI 
Christ 
quinci 
Geosc 
Harris 
and e 
189-1