The boundary transformation from the local coordinate systan 
to the corresponding line and column number of the landsat subsence, 
is performed by the programme which will make an affine transforma- 
tion. 
3/ Data from a mask which represents training areas by means 
of different thematic codes. 
B. The histogramme and the symbol coded picture of the training 
area. The programme can display them either on the terminal 
Screen or on the line-printer. 
C. When given thresholds, the pixels, with digital levels excee- 
ding the thresholds, are deleted from the symbolic picture 
and will not be included in the calculation of the statistics 
of the training area. 
D. Calculating statistical values as averages, standard devia- 
tions, covarience matrices and correlation matrices of the 
training areas. 
At last, the programme will give the user the possibility to 
combine the statistical values of the training arcas extracted 
from either optional files or the terminal, and to delete some 
statistical values which the user does not require. : 
the user can store all the statistical values of the training 
areas onto a disk file. 
5. Supervised Classification 
Classification is based on various supervised classification 
techniques that require refrence signatures of targets represented 
by training areas on the ground. 
The EPDCS system provides the user with three different Super- 
vised classificution methods-maximum likelihood classification, 
minimum distance classification and table look-up classification. 
A user can select different training areas, channels and thresholds 
for the classification, using data stored on the same file. The 
classified image on the paper with a symbolic code representing 
different classes are produced by line-printer. Classified colour 
or black-white image on the film will be produced by rotating drum 
recorder. 
6. Un-Supervised classification 
The supervised technique is associated with high variability of 
spectral signatures. When so, it is difficult to set up an opera- 
tional library of refenence signatures. You have to obtain the re- 
ference signatures directly from training areas. Even in this prac- 
tice, the training areas selected by the user are not objective. The 
unsupervised classification method avoids the above mentioned diffi- 
culty by not requiring the reference signatures in the data proces- 
ing phase. After the data processing the identification is done by 
pixelwise testing of the class assignment. The EPDCS system has 
three programmes for unsupervised classification. The first cluster- 
ing algorithm is used to group pixels with similar spectral charac- 
teristics. The second clustering algorithm is used to group pixels 
by means of the local textural parameter which was proposed by H. 
BEGUIN, H,Do TU and J. WILMET. The third algorithm also consists in 
combining spectral information with spatial information. Firstly the 
image is separated into the "small regions", pixels of which must 
be neighbours and must have very similar spectral values. Then, the 
unsupervised classification based on the "small regions" is per- 
formed according to spectral mean value and deviation of the regions 
  
  
   
   
  
  
  
  
    
  
  
  
  
  
  
  
  
   
    
   
  
   
  
  
   
   
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
    
  
he re 
vised clas 
Spectral | 
having sin 
ers from % 
formation 
is more us 
assignment 
processing 
within an 
sent diffe 
8. Accurao 
In ord 
truth, we 
and map by 
7. Ciscanin 
m 
"ing paramo 
Kk * Dif 
The ma 
fied, by m 
which have 
bols repre 
B. Con 
The ta 
es. 
C. Ovo 
D. Map 
B. Ove 
' ACKNOW] 
I am v 
Mr, Gôran ; 
University 
ideas and