Table 3: The error matrix of LVQ-classification. 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Class 1 2 3 4 5 6 7 Y, 
1 2550 110 124 0 235 62 171 3252 
2 31 8583 24 26 192 0 4 8860 
3 40 22 2816 0 145 21 104 3148 
4 0 23 0 4164 1 0 0 4188 
5 221. 355 199 4 1777 8 70 2634 
6 9 0 26 0 22 21 13 91 
7 327 9 452 0 132 101 1032 2053 
X 3178 9102 3641 4194 2504 213 1394 
Table 4: The error matrix of MRF-classification. 
Class 1 2 3 4 5 6 7 2 
1 2702 57 102 0 229 57 104 3251 
2 26 8767 0 10 103 0 1 8907 
3 26 14 3000 0 125 19 75 3259 
4 0 5 0 4183 0 0 0 4188 
5 113 252 139 1 1922 6 52 2485 
6 3 0 4 0 7 16 3 33 
7 308 7 396 0 118 115 1159 2103 
Y 3178 9102 3641 4194 2504 213 1394 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Table 5: The producer’s and user’s accuracies of the LVQ- and MRF-classifications. 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Class LVQ, producer's acc. LVQ, user's acc. MRF, producer’s acc. | MRF, user’s acc. 
1 80.2% 78.4% 85.0% 83.1% 
2 94.3% 96.9% 96.3% 98.4% 
3 77.3% 89.4% 82.4% 92.1% 
4 99.3% 99.4% 99.7% 99.9% 
5 71.0% 67.5% 76.8% 77.3% 
6 9.9% 23.1% 7.5% 48.5% 
7 74.0% 50.3% 83.1% 55.1% 
572 Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
So: 
(GC 
pe: 
Sce 
the 
vec 
cl) 
tre 
the 
pat