Nas 
airs 
f a 
del 
sult 
| of 
fied 
ates 
phic 
3PS 
ere 
ble, 
the 
rol 
one 
cted 
ned 
odel 
ted, 
ined 
odel 
co- 
not 
"on 
LED 
was. 
n of 
A is 
phic 
itour 
cale 
,000 
(dpi) 
ation 
cale 
ters 
34) 
* Computation of a 20 meters grid HEIGHT DATA 
file by interpolating contour lines and spot heights. 
The above derived DTM was mosaiced with the existed 
SPOT STEREOPAIRS DTM in order to complete the 
HEIGHT OATABASE. 
The absolute accuracy of the SPOT digital elevation 
model depends also on the scale of the maps used for 
the definition of the ground control points. SPOT digital 
elevation model was registered from 1:50,000 scale 
topographic maps and respectively has 30 meters plane 
accuracy and 10 meters height accuracy. About 2000 
points were checked, distributed in the whole country and 
the errors in height are summarised at the following table: 
= EZ M rt rte que pa sta 
xe ET E Sor MELLE MEC 2 EC dv OT 
e Rh. P URP rentrent I Veloso EN 
  
  
Table 2 : Errors in height from 2000 checked points 
Finally, using the above produced DTM the SPOT-P, 
SPOT-XS and LANDSAT TM were orthorectified. 
3.2 Clutter Database 
The CLUTTER DATABASE was extracted from 
multispectral satellite orthoimages, SPOT XS and 
LANDSAT TM (Fig.2). by classification and photo- 
interpretation techniques. 
CC] SPOT images 
M LANDSAT images 
  
Figure 2 : Clutter Data 
The Feature Space (ERDAS Imagine 8.2) decision rule 
was used for the classification of the multispectral 
images. Feature space image were used to define the 
training sample. The advantages of this method over the 
917 
  
    
EE Advantage sas 
traditional ones are . that feature space is a non- 
parametric signature, the decisions made in the 
classification process have no dependency on the 
statistics on the pixel and helps. to improve classification 
accuracy's for the non-normal classes. 
e »gixel$ in C215 © 
a = 3ixelsin cas: 2 
* = Qixals iv Cass 5 
Band 8 
shades hie value 
  
  
  
  
8and A 
daca Me values 
Figure 3 
The classified classes were : 
e forests 
e waters 
e Open areas 
E urban 
e suburban 
Urban and suburban classes were extracted from the 
photo-interpretation of panchromatic and multispectrai 
images. 
The advantages and disadvantages of using feature 
space signatures can be briefly presented at the table 
below: 
c 
Sage) 
Helpful for the first pass, | Feature Space image may 
broad classification. be difficult to interpret. 
  
An accurate way to classify 
classes with non-normal 
distribution. 
  
Features may be more 
visually identifiable, which 
can help discriminate 
between classes spectrally 
similar and it is hard to 
differentiate with 
parametric information. 
  
  
Feature space method is 
very fast. 
  
  
  
Table 3 : Advantages of using Feature Space Signatures 
3.2 Vector Map Data 
Vector Map Data were extracted from the panchromatic 
and multispectral orthoimages by interactive on screen 
digitisation techniques. 
The following planimetric features were obtained: 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996