with the small area categories. On the method. the lowest for the elemene ratio 
  
contrary, in the case of the maximum matching method. 
percentage method, the clusters are 
hardly labeled with large area catego- (4) From the view point of practical use, 
ries. These are general problems for the maximum number method,the maximum 
these two methods, because usually the percentage method and the element ratio 
area of each category is not the same. matching method are easy to execute be- 
cause they don't need the spectral fea- 
(2) The classification accuracies(area ture, but the classification accuracies 
weighted mean) were the hightest for the (simple mean) of three methods are lower 
element ratio matching method but detail than that of the minimum distance method. 
information such as river or road was 
lost as shown in Fig. 6 because that (5) Among the methods which were evaluat- 
cateogorization was performed on each ed, the minimum distance method showed 
local region. best result. In this method, obtained 
result is almost the same with a super- 
(3) The classification accuracies(simple vised method. Theoretically, this method 
mean) were the hightest for supervised also needs less number of pixels for TCA 
method, second, the minimum distance compared to other 3 methods because the 
geometrical information of TCA is not 
used. 
category area and occupation rate categorized by | categorized by 
of cluster k in that category area maximum number | max. percentage 
method method 
  
> 
œ 
© 
  
  
  
Clas 
k 
  
  
  
  
  
  
  
  
  
  
  
> 
> 
  
  
  
  
  
  
  
  
  
  
  
  
  
NE 
: NE 50% 30% 20% 
NE (b) case of same size of category area 
SS A 
A BC B 
C 
(a) cluster k, 
composed of ER A C 
category A,B,C 
22% 30% 80% 
(c) case of different size of area 
  
  
  
  
  
  
Fig. 1 Cluster and Categories 
141