International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
will be constructed by more elements for the data acquired at 
the same area in 2002 because whole polarization data were 
acquired for the observation. 
From the mentioned discussions, Rajski distance proposed in 
this paper is considered to be suitable as the element of feature 
vector for land-cover classification. 
6. CONCLUSIONS 
In this paper, Calculating Rajski distance from polarimetric 
SAR amplitude image data and introduce of it to land-cover 
classification as element of feature vector were proposed. To 
obtain Rajski distance, gray level co-occurrence matrix was 
constructed using two amplitude image data and joint entropy 
and conditional information contents were calculated using the 
matrix. Actual polarimetric SAR data, SIR-C and Pi-SAR, were 
applied to proposed algorithm. To introduce Rajski distance as 
elements of feature vector, the distance was quantized to gray 
scale image like other amplitude images. In Rajski distance 
images, some characteristic properties were appeared for water 
area and vegetation. Average classification accuracies for whole 
data were improved when extended feature vectors were 
applied actually to land-cover classification. The size of sub 
area extracted to construct GLCM for the area that has narrow 
spatial range needed correction. Optimize of size of sub area 
and introduction of other parameter obtained from GLCM will 
be needed to advance. 
ACKNOWLEDGEMENT 
The authors would like to thank Dr. Masaharu Fujita, Professor 
of Tokyo Metropolitan Institute of Technology, Dr. Seiho 
Uratsuka, Group Leader of Environment Information 
Technology Group, Applied Research and Standard Dept., 
National Institute of Information and Communications 
Technology, offered the data of SIR-C and Pi-SAR. 
REFERENCES 
Yueh, H. A., et al., 1988. Bayes Classification of Terrain Cover 
Using Normalized Polarimetric Data. J. Geophysical Research, 
93(B12), pp.15261-15267. 
Ito, Y.. and Omatsu, S. 1997. Polarimetric SAR Data 
Classification Using Neural Networks. J. Japan Society of 
Photogrammetry and Remote Sensing, 36(3), pp.13-22. 
Zyl J. J, 1989. Unsupervised Classification of Scattering 
Behavior Using Radar Polarimetry Data. /EEE Trans. 
Geoscience and Remote Sensing, 35(1), pp.68-78. 
Cloude, S. R. and Pottier, E., 1997. An Entropy Based 
Classification Scheme for Land Applications of Polarimetric 
SAR. IEEE Trans. Geoscience and Remote Sensing, 35(1), 
pp.68-78. 
Isomichi, Y., 1980. Information Theory. Corona publishing co., 
Itd., Tokyo Japan, pp.6-34 
Haralick, R. M., et al, 1973. Textural Features for Image 
Classification. [EEE Trans. Systems, Man, and Cybernetics, 
SMC-3(6), pp.610-621. 
26 
Yamada, T. and Hoshi, T., 2002a. Expansion of Feature Vector 
Introduced Rajski Distance for Land-cover Classification using 
Polarimetric SAR Image Data. J. Japan 
Photogrammetry and Remote Sensing, 41(6), pp.14-26. 
Yamada, T. and Hoshi, T., 2002b. Application of Rajski 
distance to pseudocolor synthesis for recognition of dual 
polarization SAR image data. Proc. 35th Conference of The 
Remote Sensing Society of Japan, pp.127-128. 
Society of 
KEY 
ABS 
In th 
to be 
Gaus 
estin 
base 
Quar 
noise 
Imag 
imag 
whic] 
phen 
const 
returi 
prodi 
appe: 
as mi 
Many 
apprc 
the s 
the ir 
noise 
is cal 
apert 
proce 
dime 
imag 
the s 
differ 
reduc 
smoo 
Thes: 
withi 
new 
comp 
statis 
bette 
imag 
Both 
the e 
imagi 
be b: 
detail 
Gene 
accor 
homc 
point