and marshes and some old roads and clear-cuts. The elevation 
ranges from around 250 to 475 m. above sea level. 
The SPOT panchromatic scene was recorded on August 13, 
1995 and has been radiometrically and geometrically corrected 
(to the Swedish national grid) The registration angle was 5.24 
degrees towards east (figure 2). 
  
Figure 2. SPOT Pan image of the study area. Forest and water 
bodies are dark, while clear-cuts of different age, roads and 
wetlands are grey-white. 
The topographic map is available in digital form from NLS. It is 
based on aerial photos from 1992 and field work from 1993. 
The Geographic Resource Analysis Support System (GRASS; 
US CERL, 1993) in conjunction with programs written in C 
have been used in the study. GRASS is a raster based 
geographic information system which includes some basic 
image analysis capabilities. 
3.2 Identification of revision objects 
The identification of revision objects is based on characteristics 
which describe the objects in terms of spectral, spatial and 
contextual properties. The object descriptions are translated into 
rules which serve as the base for identification. The rules are of 
the form IF «condition» THEN «conclusion», where the 
condition often includes vague statements such as "the spectral 
reflectance is high", and fuzzy logic is applied in order to 
translate the statements into mathematical terms. 
Example roads 
In addition to high spectral reflectance the shape of a road is 
important to separate it from other revision objects. These two 
characteristics were used to construct a rule which describes the 
revision object road. Each important characteristic, or factor, is 
seen as a class to which the membership for each pixel in the 
image should be evaluated. Additionally, the pixels are 
compared to the existing map to avoid detection of previously 
mapped roads. The following rule was used to identify roads: 
IF high spectral reflectance 
ANDIF line, shaped and thin 
ANDIFNOT existing. road 
THEN new. road. 
Before it is possible to determine what “high spectral 
reflectance” is, the typical spectral reflectance of forest in this 
image and this area must be known. The median value for the 
image was computed as an approximation of the typical spectral 
characteristics of forest. No other land use type besides forest 
exists within the test area, and the approximation is based on 
the fact that the forested area is considerably larger than the area 
covered by water, clear-cuts, roads, rock outcrops etc. 
Membership functions for the two first factors in the rule were 
calculated. Two map layers showing membership values in each 
pixel for high spectral reflectance (figure 3) and 
line shaped and thin (figure 4) respectively, were created . 
The membership functions were calculated in the following 
manner: 
1 forx > max 
: x - min ; 
HA (high spectral reflectance) = formin « x « max 
max - min 
0 forx « min 
where min = (median + 0.5 * standard deviation) 
max = (median + 2 * standard deviation) 
(cf. Wang 1994). 
The thresholds (min and max), expressed in standard deviation 
from median in the formula, were estimated based on grey 
values in recent (unmapped) clear-cuts. Small changes of the 
factors used for multiplication of standard deviation (0.5 and 2) 
did not significantly affect the membership values. 
All deviating pixels (i.e. the pixels with a membership function 
larger than 0) in the created layer were considered as candidates 
for identification both for clear-cuts and for roads. 
  
  
     
  
   
  
S. d un of a , "T. : = 
Figure 3. Membership values for fuzzy statement “high spectral 
reflectance”. Dark areas have membership values close to 1. 
Bright areas have values close to 0. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996 
  
The 
the 
enh: 
con: 
Foui 
filte 
a ra 
was 
  
Figu 
and t 
dark 
À s 
line. 
with 
the s 
and 
value 
oper: 
(high 
fuzz 
value 
each 
the o 
show 
Altm 
map 
Exar 
The | 
same