stanbul 2004 
(6) 
obtained as 
/Xq) for the 
| is obtained 
nstructed by 
ls of single 
ribed. Main 
jyllows. 
-occurrence 
Note the size 
probability 
area of tiny 
) large size. 
Lb. dn the 
y for the 
images are 
constructed. 
sformed to 
dint entropy 
d using the 
distance is 
entropy and 
ixel Value: 
ue that set a 
responds to 
pixels with 
ski distance 
lustrated in 
listance 
ige 
  
  
  
  
  
  
  
onstruct 
e matrix 
  
mage 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
3. APPLICATION OF POLARIMETRIC SAR DATA 
The proposed technique in this paper was applied to land-cover 
classification using two types of polarimetric SAR data: SIR-C 
and Pi-SAR. L-band and C-band data observed by SIR-C at 
Sarobetsu site and Kashima site in 1994 were applied, and X- 
band data observed by Pi-SAR at Hitachi site in 1998 and 2002 
were applied. Examples of Rajski distance image created by 
each observation site in Figure 2, 3, and 4. The created images 
for SIR-C Sarobetsu site data are shown in Figure 2, for SIR-C 
Kashima site data are shown in Figure 3, and for Pi-SAR 
Hitachi site data are shown in Figure 4, respectively. In this 
experiments, sub areas were extracted with the size of 9 x 9 
pixels when Rajski distance was calculated. 
As the common property, the tendencies that Rajski distance is 
higher in water area and lower in vegetation are shown in all 
images. In water area, pixel values that exist in sub area are 
different slightly although these values are quite similar. So that 
it can consider that Rajski distance is higher because mutual 
information contents was lower due to narrow deviation of pixel 
value. On the other hand, the property of Rajski distance in 
vegetation can be described relating with microwave scattering 
model. It is known widely that notable diffusion scattering 
appears in this area. In this case, helix components are included 
largely according to basic scattering model. Rajski distance is 
lower due to similar pixel value increases equally in each 
amplitude image because the elements in scattering matrix for 
helix have same value. This reason has been confirmed by 
simulation. 
To introduce Rajski distance to land-cover classification, 
feature vector is constructed with pixel value of Rajski distance 
images and amplitude images. 
  
0 E 1 
Figure 2. Example of Rajski distance image for SIR-C 
Sarobetsu site C-band data (1994) 
  
Figure 3. Example of Rajski distance image for SIR-C 
Kashima site C-band data (1994) 
  
  
(b) HH-HV polarization 
BE |i 
Figure 4. Example of Rajski distance image for Pi-SAR 
Hitachi site X-band data (2002)