The results show that accounting for intra and inter class 
variability becomes easier by using the object oriented multi- 
resolution segmentation and classification approach. 
Similarly, use of contextual information such as occurrence 
of banana plantations along River Krishna by defining rules 
helped in better classification of the study area. Analysis of 
classification results by per pixel classifier shows that (i) it is 
difficult and often impractical to account for all the inter and 
intra class variability by defining training areas at pixel level 
and (ii) per pixel classifier outputs contain misclassified 
pixels along the class boundaries. Results obtained by object 
oriented multi-resolution segmentation and classification 
reveal that (i) The study shows that segmentation of remote 
sensing data prior to its classification generates 
homogeneous segments or image objects and enables to 
account for all variations, (ii) the approach is less time taking 
and noise free at all resolutions and scales. It was seen that 
by using the contextual and hierarchical (multi-resolution) 
information misclassification of objects with similar spectral 
response patterns could be significantly reduced. 
3. CONCLUSIONS 
In the present study object oriented multi-resolution 
segmentation and classification approach has been applied to 
the IRS-1D LISS-III data acquired over two test sites. The 
results have been compared with those obtained by standard 
maximum likelihood per pixel classifier. Results of the 
adopted approach reveal that inter and intra class variability 
associated with high resolution remote sensing data could be 
accounted for by segmentation in the first step, followed by 
classification. Performance of the adopted approach is 
superior to that of per pixel classifier. The approach is 
flexible with the options of (i) interactive as well as 
automatic classifications, (ii) using class hierarchy and 
contextual information, and (iii) extracting information 
residing at different resolutions and scales. However, multi- 
resolution segmentation is based on certain heuristics and 
IAPRS & SIS, Vol.34, Part 7, "Resource and Environmental Monitoring", Hyderabad, India,2002 
data dependent. Sufficient care needs to be exercised during 
image segmentation to realize acceptable accuracy of 
classification. 
ACKNOWLEDGEMENTS 
Authors area grateful to the Director and the Deputy Director 
(RS&GIS), NRSA for supporting the work. Authors wish to thank 
the support extended by Mr. Jagadeesh, formerly with ND Soft 
Spatial Systems Pvt. Ltd., during the initial stages of the work. 
REFERENCES 
Chaudhuri, B. and N. Sarkar, 1995, Texture segmentation using 
fractal dimension, IEEE Transactions on Pattern Analysis and 
Machine Intelligence. Vol. 17, 1. pp. 72-77 
Haralick, R., Shanmugan, K and I. Dinstein, 1973, Textural 
features for image classification, IEEE Transactions on Systems, 
Man and Cybernetics. Vol. 3, 1, pp. 610-621 
Manjunath, B. Et R Chellappa, 1991, Unsupervised texture 
segmentation using Markov random field models, IEEE 
Transactions on Pattern Analysis and Machine Intelligence. Vol. 
13, pp.478-482. 
Mather, P.M., 1990, Theoretical problems in image classification, 
pp. 127-135, Applications of remote sensing in Agriculture, edited 
by Stevenson, M.D., and J.A.Clark, Butterworths, London 
Mao, J and A. Jain, 1992, Texture classification and segmentation 
using multiresolution simultaneous autoregressive models, Pattern 
Recognition, Vol. 25, pp. 173-188. 
Salari, E. and Z. Ling, 1995, Texture Segmentation using 
hierarchical Wavelet Decomposition, Pattern Recognition, Vol. 
28, 12, pp. 1819-1824. 
    
2) 
KE 
ABS 
oveı 
the 
seri 
rem 
the | 
Incr 
has 
reso 
Syst 
cali 
are 
coai 
coa: 
rela 
as S 
Int 
pres 
gair 
The 
sucl 
usin 
gen 
amc 
Met 
ban 
mat 
The 
radi 
com 
afte 
slov 
80s 
Thi: 
nois 
LA] 
Thi: 
dire 
The 
amp 
16-] 
line