International Archives of the Photogrammetry, Remote Sensing 
Table 3. Accuracy results for classified image from pixel-based cl 
     
and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
  
  
  
assifications and object-oriented image analysis. 
  
  
  
  
  
  
  
  
  
  
   
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Paralellpiped Classification Minimum Distance Classification Maximum-Likelihood Object-Oriented Classification. 
Classification : 
Class Name Producer User's Kappa Producer User's Kappa Producer's User's Kappa Producer's User's Kappa 
s Accuracy Statisti S Accurac Statistic Accuracy Accurac Statistic Accuracy Accurac | Statistic 
Accuracy % c Accuracy y 96 % y Yo % y Yo 
% % 
92.632 100.000 1.0000 94.737 100.00 1.000 94.737 100.00 1.0000 100.00 100.00 1.0000 
dam lake 0.000 0.000 0.0000 0.000 0.000 0.000 0.000 0.000 0.0000 100.00 100.00 1.0000 
settlement areas 11.111 66.667 0.6486 16.667 100.00 1.000 11.111 66.667 0.6486 50.00 75.00 0.6670 
dense forrest 94.792 65.942 0.5307 98.958 56.548 0.401 93.750 67.164 0.5475 100.00 100.00 1.0000 
open areas 15.094 61.538 0.5467 9.434 62.500 0.558 16.981 47.368 0.3798 80.00 66.70 0.6280 
coal waste 12.500 100.000 1.0000 12.500 11.111 0.090 12.500 33.333 0.3177 80.00 50.00 0.4420 
wood land 45.000 38.710 0.2055 31.250 33.211 0.160 52.500 41.584 0.2428 91.70 91.70 0.8890 
  
  
  
  
  
Table 4. Kappa and overall accuracy values for pixel-based classifications and object-oriented technique. 
  
  
  
  
  
  
  
  
  
  
Accuracy Statistics Paralellpiped Minimum Distance Maximum-Likelihood Object-Oriented 
Classification (%) Classification (9) Classification (76) Classification (96) 
Overall Accuracy 64.571 62.571 66.857 81.30 
Overall Kappa Statistic 0.532 0.499 0.558 0.766 
  
  
  
  
  
  
  
  
However, object-oriented classification produced more 
accurate results. The reason for this is that the compactness 
of the segments. Thus, kappa and the overall accuracy are 
much better. In table 4, while overall accuracy was 81.30 for 
object-based segmentation, it was only 66.86 for the best 
classical classification technique of maximum likelihood. For 
kappa values, some trends occurred and it is 0.77 for object- 
based image analysis. 
5. CONCLUSION 
In this paper, new object-oriented image analysis technique 
has been compared with the classical and the well-known 
image classification methods using Landsat-7 ETM image of 
Zonguldak testfield. In the implementation of the tests, 
paralelepiped minimum-distance and the maximum- 
likelihood approaches are taken as pixel-based methods. 
Their capacity with used Lansdat image has been analysed 
based on the ground truth materials over the interest area. 
Howver, on the other hand, eCognition software for object- 
based  classfication works in hierarchy, first with 
segmentation, then the fuzzy classication. Detailed accuracy 
results were obtained as error matrices and they show that the 
object-based image anaysis is far beyond the classical 
methods in terms of accurate classification of the objects. 
6. ACKNOWLEDGEMENTS 
This work was carried out under the projects supported by 
TUBITAK (Turkey) - JULICH (Germany) cooperation with 
a code no. 101Y090 and Karaelmas University project no, 
2001/45-01. 
7. REFERENCES 
Benz, U. C., Hofmann, P., Willhauck, G., Lingenfelder, I. 
2003. Multi-resolution, object-oriented 
and Heynen, M. 
GIS-ready 
fuzzy analysis of remote sensing data for 
information, ISPRS Journal of Photogrammetry & Remote 
Sensing, 58 (2004) pp. 239-258 
eCognition User Guide 3. 2003. Definiens Imaging, pp. 3.2- 
108 
Lillesand, M.,Tt., Kiefer, W., R. (1994). Remote Sensing and 
Image Interpretation, Third Edition, John Wiley & Sons, 
Inc., New York, 750 pp. 
i 
KEY 
ABS 
Duri 
Offic 
Inter 
The 
ment 
The 
the r: 
of 1: 
by R 
meta. 
The « 
Arcll 
interí 
The : 
(WFS 
The s 
to be 
Durir 
Rom: 
large 
As lc 
series 
z t 
M m [no fI 085 M ^ 
The t 
paper 
presen 
paper 
geolog 
Inforn