tiation uses the recognition result obtained in the previous 
step. 
9. INTERFACE BETWEEN ANALYSIS SYSTEM 
AND GIS 
The facts extracted by the map interpretation are stored in 
the instance base. For further use of the corresponding 
information the data has to be converted to data structures 
specified by the INFO-Database of the GIS. This conver- 
sion has to be done by the module data conversion (cf. 
Fig. 1). 
10. IMPLEMENTATION AND RESULTS 
Our system for knowledge-directed analysis of maps con- 
tains currently the basic ideas and methods presented here. 
So far the module conflict solution, described in section 8, 
is not yet implemented, however this will be done in near 
future. Algorithms for processing raster data are imple- 
mented using High-C (from Metaware). The knowledge- 
directed system is realized by the object-oriented program- 
ming language and development environment Smalltalk- 
80 (Goldberg et al., 1989). 486-PCs are serving as host 
computers. 
Fig. 11 shows a test scene scanned from a topographic map 
of scale 1:5 000 containing the colors black and brown. 
Figs. 12 and 13 show the two color layers BROWN and 
BLACK that have been separated from the original RGB- 
image of Fig. 11 using the methods described in section 
4.1. 
  
  
  
  
  
  
  
  
  
  
Fig. 11: Example of a topographic map scene (original 
size is 70x44 mm resp. 1104x700 pels). 
/ I -— — * 
| “m= NN. 
| Z ^N j \ 
| / => A / 
| 12,0 Z 
/ / —— 2 = na tm À à v 
n z nt 
| / ST Te. = A 
4 25 / € t 
ACER ^ IE 
a ^ : —9-" 
A # rugs ve to e — pen 
ut -— — rM 
d ^ rd s [ = 
et. ed bh ve pe 
Pad « x J 
v y €——— rn aes i 
Fig. 12: Color layer BROWN separated from image shown 
in Fig. 11. 
662 
  
  
  
  
  
  
Fig. 13: Color Layer BLACK separated from image shown 
in Fig. 11. 
For first experiments the model represents concepts for 
interpretation of text symbol, coniferous tree, deciduous 
tree, bush, heath, road section, crossing, contour line (Im 
interval) and contour line (5m interval). The correspond- 
ing interpretation result for the above example is shown in 
Fig. 14. The image parts without marking symbols are not 
recognized in the present system. 
  
  
» à x 
e^ re Jpn il s 
e B 
A o Sa 9 1 ! 
9 + PT LI p E = $ =? 
© deciduous tree A coniferous tree À bush EX heath S recognized symbol 
  
  
  
contour line = = Contour line road section ; 
7^7" Sm interval 1m interval (parallel lines) $a crossing 
Fig. 14: Interpretation result for map shown in Fig. 11. 
11. SUMMARY AND CONCLUSIONS 
An overview of an image analysis system for interpretation 
of topographic maps was presented. Methods for raster 
data processing, knowledge organization and knowledge 
use were discussed. The main ideas of raster data pro- 
cessing are scanning using a 24-Bit-RGB-scanner, separa- 
tion of color layers, raster symbol and raster object recog- 
nition and vectorization. The principles of knowledge- 
directed interpretation are those of prototypes (concepts) 
as the basic representation building block, generalization, 
and aggregation as interacting abstraction mechanism. Fi- 
nally the capabilities of the system are demonstrated on a 
map scene. Further work will be directed towards the 
conflict solution module and the improvement of the al- 
ready existing methods. Eventually, the system will be 
tested for more map scenes. Furthermore, it must be inves- 
tigated how more complex raster processing steps for ex- 
traction of structure primitives (Ebi et al., 1991) may 
improve the overall system performance.