Semi-Automatic Revision of Topographic Maps 
from Satellite Imagery 
Rune Solberg 
Norwegian Computing Center 
P.O. Box 114 Blindern 
N-0314 Oslo 
Norway 
Phone: + 47 2 453500 
Fax: +47 2 697660 
E-mail: Rune.Solberg@nr.no 
Abstract 
The Norwegian Computing Center has in collaboration with 
The Norwegian Mapping Authority developed a semi- 
automatic prototype system for revision of some features in 
1:50.000 topographic maps based on pattern recognition in 
SPOT satellite imagery. The system generates a proposal for 
new roads and new urban areas. Change detection is 
performed in two main steps: In the first step the unrevised 
digital map is used for detection of unchanged features in the 
imagery. In the second step, the features recognized in the 
first step are removed from the segmented data, and the new 
objects are detected from the rest of the features. The revision 
result is used for updating the map data base for production 
of "satellite revised maps". A satellite revised map is 
proposed as an intermediate map between the aerial photo 
based map revisions each 10th to 15th year. 
1. Introduction 
With a map revision cycle of 10-20 years for the main 
topographic map series (1:50.000) in Norway, the mapping 
organizations have been interested in trying to develop a 
semi-automatic map revision system based on satellite 
imagery. Our idea has not been to replace airphoto based map 
revision, but to supplement it with satellite imagery based 
map revision between the airphoto based revisions. The 
product could be "satellite revised" maps, published two to 
four times more often, with changes based on the satellite 
data printed in e.g. other colours to indicate the greater 
uncertainty with these changes. 
The project started in 1988 as a joint venture between the 
Norwegian Computing Center (NR), the Norwegian 
Mapping Authority (NMA), the Ministry of Defence's 
Cartographic Department, and the Royal Norwegian Council 
for Scientific and Industrial Research (NTNF). The main part 
of the development has been performed at NR complemented 
with testing and some hardware dependent development at 
NMA. Revision of the road network and urban areas have 
been given preference since these features change most 
rapidly. 
2. System Overview 
The satellite imagery interpretation is divided into two main 
functions, map-guided detection of old features and non- 
guided interpretation of new features. The results from the 
first part are used to remove the old features from the image 
to determine which features can be new. The result from the 
549 
second part can and should be visually controlled and edited 
before the features are converted to vector format to update 
the database. 
The model for the general system design is shown in figure 
1. On the far left, satellite images are stored in a database 
from which they are read into the preprocessing module. It is 
assumed that the images already have been radiometrically 
and geometrically corrected. The main function of the 
preprocessing module is to use segmentation methods for 
extracting structural objects, such as segments of roads or 
built-up areas, from the satellite imagery. In order to guide 
the segmentation process digital map data is extracted from 
the cartographic database. The structural objects which are 
recognized are normally line-like or region-like in shape 
according to the type of information which one wants to 
obtain from the satellite image. For example, if one is 
interested in updating roads, then one would use a 
, segmentation method which extract line features, whereas for 
updating built-up areas, a method which extracts regions 
would be used 
The core of the system is the image interpretation module. 
The basic information used in the module is (1) the result of 
the segmentation, (2) the original image, and (3) the old map 
represented in a raster format. The main function of the 
module is to use this basic information in order to identify 
new and changed map features, and to make proposals for 
revision. During execution, the user will see on the screen 
which map features the image interpretation module has 
identified as those which already exist on the old map, and 
those which are different from those on the old map. As 
features are identified, they are drawn on the screen, and 
different colours are used to indicate whether a feature is 
unchanged, changed, removed, or new. Because of the 
complexity of satellite imagery, it is important that the system 
indicates where uncertainties have arisen during the image 
interpretation phase. For example, the system may not be 
able to identify with certainty a particular linear feature, if the 
feature is in a noisy part of the image. These indicators can 
then help the operator when he is checking the result of the 
image interpretation. 
The map manipulation module is connected to a cartographic 
database where map data are represented in a vector format. 
This module takes the map data from the area of interest and 
converts them to a raster format before sending the map to the 
image interpretation module. After image interpretation the 
proposals for map revision are returned to the map module so 
that the operator can check the result and make changes if