regular, 
e them, 
of their 
nd poor 
afficient 
ognised 
e 
lows on 
1 easier. 
are well 
d in two 
nsity of 
gether a 
Ises, are 
e can be 
classes 
or not. 
ey make 
ment of 
512x512 
Distance 
; method 
re mixed 
Water is 
'rcentage 
es which 
can also 
es which 
is regular 
flats and 
n road is 
ig to the 
is way a 
example 
  
  
  
  
    
Figure 4. Results of classification by means of the other 
methods. a) K-Means, b) ISODATA, c) Simple One Pass 
Clustering, d) Minimum Distance to Mean . The bands used: 
Red, Blue, Green, Near IR, (1m resolution, 512x512 
pixels). According to the Figures, these methods are less 
efficient compared with Fuzzy-C and Maximum likelihood. 
In color. 
The source of 4m resolution and an example of the 
classification results by all the methods for 4m resolution 
are given on Fig. 5 and Fig 6. 
tT 
Figure 5. RGB visual bands image ( 4m ). In color. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
  
Figure 6. Results of classification by all methods, applied on 
4m resolution image. The bands used: Red, Green, Blue, 
Near IR. a) K-Means, b) ISODATA, c) Simple One Pass 
Clustering, d) Fuzzy-C, e) Maximum Likelihood, f) 
Minimum Distance to Mean. In color. 
Compared with sources and classification of the Im 
resolution images at the most efficient methods these results 
for 4m resolution are not satisfactory. 
4. EVALUATING CLASSIFICATION - ACCURACY 
ASSESSMENT 
Evaluation of resolution influence on the classification 
efficiency and methods is discussed in the previous section 
which came to the conclusion that only two methods Fuzzy- 
C Means and Maximum Likelihood should be considered on 
the lm resolution images. Evaluation of classification 
efficiency is made at three control areas including the most 
significant objects Fig. 7, Fig. 8 and Fig. 9. Table 1. shows 
errors of the classification. 
659