when the window size continues to increase, the classification 
accuracy increases but smoothly and slowly until it reaches to 
the highest value when the window size increases again; 
afterwards, with the increase of the window size, the 
classification accuracy decreases. 
So we can conclude that for a certain ground object, the 
estimated scale can be used as the suboptimal selection for the 
processing window size during the texture extraction. Examples 
are given in Figure 7. 
REFERENCES 
B.B. Chaudhuri and N. Sarkar, 1995. Texture segmentation 
using fractal dimension. IEEE Transactions on Pattern Analysis 
and Machine Intelligence, vol. 17, no.l, pp.72-77. 
B.S. Manjunath and R. Chellappa, 1991. A Note on 
unsupervised texture segmentation. IEEE Trans. Pattern Anal. 
Mach. Intel!, vol. 13, no. 5, pp. 478-483. 
Growth Rate of Classification Ac curacy 
0.7 
0.6 
0.5 
0.4 
0.3 
0.2 
0.1 
r-rl 
1 
g,.. -“-(-f'— 
0.1 
1 2 
3 4 5 6 7 8 9 
10 11 12 13 14 
15 16 17 
□ Cor.forVeg. 
Figure 7. An example of growth rate of classification accuracy 
on the change of window size for vegetation 
3.3 Multi-scale Texture Feature Classification 
According to our experiments, we find that different texture 
feature has different presentation for a certain ground object. 
Correlation image is suitable for vegetation recognition, 
Entropy image is suitable for residential area classification, and 
Stand Deviation image has the similar interpretation abilty for 
vegetation, residential area, and water body. The optimum 
multi-scale features selected can give an improved classification 
performance compared with a single-scale classification with 
the given result in Table 2. 4 
Multi-scale 
COR 11/ENT 13/SD 21 
Single-scale: 11 
Class Type 
Prod. Acc. 
User Acc. 
Prod. Acc. 
User Acc. 
Water Body 
42.80% 
53.03% 
35.84% 
56.11% 
Vegetation 
85.31% 
64.32% 
87.26% 
60.28% 
Residential Area 
56.22% 
72.57% 
51.04% 
73.71% 
Overall Accuracy 
66.8931% 
63.6437% 
Table 2. Classification accuracy for different scale 
4. CONCLUSION 
C. M. Pun and M.C.Lee, 2003. Log-polar wavelet energy 
signatures for rotation and scale invariant texture classification. 
IEEE Trans. Pattern Anal. Mach. Intel!, vol. 25, no. 5, pp. 590- 
603. 
D. A.Clausi, 2000. Comparison and fusion of co-occurrence, 
Gabor and MRF texture features for classification of SAR sea- 
ice imagery. Atmosphere-Ocean, vol. 39, no. 3, pp. 183-194. 
D.A. Clausi, 2002. An analysis of co-occurrence texture 
statistics as a function of grey level quantization. Canad. J. 
Remote Sens. vol. 28, no. 1, pp. 45-62. 
D.A.Clausi and B.Yue, 2004. Comparing cooccurrence 
probabilities and Markov random fields for texture analysis. 
IEEE Trans. Geosci. Remote Sens., vol. 42, no. 1, pp. 215-228. 
D.A.Clausi and M.E.M.E.Jemigan, 2000. Establishing Gabor 
filter parameters for optimal texture separability. Pattern 
Recognit., vol. 33, no. 11, pp. 1835-1849. 
FI.Deng and D.A.Clausi, 2005. Unsupervised segmentation of 
synthetic aperture Radar sea ice imagery using a novel Markov 
random field model. IEEE Transactions on Geoscience and 
Remote Sensing, vol. 43, no. 3, pp. 528-538. 
L. K. Soh and C. Tsatsoulis, 1999. Texture analysis of SAR sea 
ice imagery using grey level co-occurrence matrices. IEEE 
Trans. Geosci. Remote Sens. vol. 37, no. 2, pp. 780-794. 
M. Tuceryan and A.K.Jain, 1993. Handbook of Pattern 
Recognition and Computer Vision, Chapter 2: Texture Analysis. 
World Scientific, Singapore. 
O.Pichler, A.Teuner, B.J.Hosticka, 1996. A comparison of 
texture feature extraction using adaptive gabor filtering, 
pyramidal and tree structured wavelet transforms. Pattern 
Recognit., vol. 29, no. 5, pp. 733-742. 
GLCP-based methods appear to be the most commonly used 
and are the most predominant method for image texture analysis 
and presentation. A multi-scale texture analysis method based 
on co-occurrence probabilities for texture analysis of single 
band and single polarization SAR imagery is presented and 
implemented in this paper. Experimental results prove that it 
can generate improved classification results compared with a 
single-scale based classification scheme. Furthermore, the semi- 
variogram model based method can estimate the scale of ground 
objects which can be further used as the suboptimal selection 
for the processing window size during the texture feature 
extraction. 
R. M. Haralick, K. Shanmugam, I. Dinstein, 1973. Textural 
features for image classification. IEEE Trans. Syst. Man Cyber. 
vol.3, no.6, pp. 610-621. 
S. E. Franklin, 2001. Remote sensing for sustainable forest 
management. Boca Raton, FLA: Lewis Publishers. 
KEYWORDS: Ta 
elet 
ABSTRACT 
Commercially availa 
geo-spatial informatif 
humanitarian missioi 
remote sensing data 
organizations with th 
radar satellite TerraS 
capabilities, offer a w 
1 
Due for launch in J 
satellite TerraSAR-X 
unprecedented quality 
for increasingly diver 
applications. Its sche 
the whole SAR progr; 
Commercial users l 
individual requirem 
independent of dayli 
and performance of 
requirements. The 
instruments of the sp 
images, day and ni] 
acquired data is the 
services, such as hi 
interpretation, topogi 
geo-spatial databases 
a wide scope of appli 
environmental plann 
exploration, regional 
response and relief, i 
applications in border 
New-quality data r 
will also offer a vas 
for instance in fore; 
oceanography. The 
surface due to plai