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
