393 
The scanning original thus prepared is scanned by means of 
a colour-scanner with a resolution of 100 micrometers. 
During the scanning process the colours contained in the 
originals are recognised and assigned to the corresponding 
pixel of the image matrix in the form of colour code 
numb e r s . 
In a further processing step at an image processing system 
the identified colour code numbers are assigned to all 
image pixels of an area enclosed by peripheries. This is 
realised by locating in the image matrix the places of the 
colour marks entered manually on the scanning original. By 
means of a N.4-neighbourhood operation this colour code 
can be assigned to all elements of the reference area. 
During this process you remark all gaps in peripheries, 
you have not corrected before. They can be closed now by 
an interactive process. 
Our objective of the processing is the selection of carto 
graphic objects of the same quality. The code relates to 
the quality. So it's very easy now to select objects of a 
certain quality. If you want to select the road net you get 
a binary image matrix in which all pixels of the road 
pattern have the value 1. Considering the line width of 
the original road signature we start a contour-fo11 owing 
algorithm and get vector data of the object contours. In 
several applications we are only interested in the coordi 
nates of the road axis, for instance. For that purpose we 
prefer algorithms which calculate the axis coordinates by 
means of the extracted vector data of the road contours. 
Also possible is the application of an image processing 
operation in form of a skeleton algorithm. But these 
algorithms are very computer-time-consuming. 
Fig.2 demonstrates a section of the planimetry representa 
tion of the test sheet and two results of feature-se1ected 
raster plotts. 
At that time single cartographic symbols - for instance 
the symbols of trigonometric points or of memorials - and 
the position of the first character of a geographic name 
are digitized for our tests on a table digitizer manually. 
An automatic pattern recognition process is planned and in 
the development. 
CONCLUSION 
The first test results demonstrate that our proposal is a 
possibly way. But at that time we have only the result of 
one test sheet. We have not enough knowledge of the 
economic advantages of this method in comparision to other 
versions of automatic digitization processes. We also need, 
for instance, some improvements in the algorithms for the 
extraction of breakpoints in a line with a polygon 
characteristic (the outlines of building signatures).