4. FROM ALGORITHMS TO PRODUCTION LINE 
The requirement of industrial data production implies 
the integration of the tools described above in a 
hardware and software environment including all 
steps of a production line from initial data input to 
sending of the final product. 
Such an environment is operated by ISTAR since 
1991. 
The hardware configuration is based on a network of 
UNIX workstations, 10 Gbytes of SCSI disk storage 
and various I/O devices (CCT, streamer and 8 mm 
cartridges tapes drives). 
The software environment is built on a raster/vector 
processing kernel with advanced visualization 
capabilities and a semi-automatic graphic editor. 
Features data are extracted from orthoimages 
produced by ISTAR altimetric production line (ref. 
Renouard). 
Since 1991, ten SPOT scenes have been processed. 
The cartographic data produced cover approximately 
30 000 km“ mainly in Europe and Middle-East 
countries. 
This first experience provides a good estimation of 
production costs: 
Land cover extraction from a geocoded SPOT scene 
(~ 3000 km?) requires about 1 man-week, whatever 
the density of features is. 
For the same input, networks extraction requires 
from 1 to 1.5 man-week, slightly depending on the 
features density. 
5. CONCLUSION 
The joint availability of high resolution raster data 
from SPOT or LANDSAT TM, and powerful 
geographic information systems (GIS), has induced 
demand for a new concept: data conversion from 
raster to planimetric vectors files fit for use in GIS. 
The new product has to preserve the same marketing 
features as original data, that is set prices and 
delays. This is possible only if a computer assistance 
is supplied. 
In order to quickly settle a computer assisted 
production line, we have designed a two level 
interactive system: 
- High level tasks, implying semantic data analysis, 
are managed by human operators. 
- Low level tasks, requiring a lot of accurate data 
processing, are done by the computer. 
The performances allow the commercial production 
of cartographic data bases. 
Future upgrades will tend to increase automation of 
networks extraction, especially in aerial images with 
higher resolution. 
6. REFERENCES 
Bretaudeau F. Une approche pour augmenter la 
séparabilité des textures. 7eme congrès de 
reconnaissance des formes et  d'intelligence 
artificielle, Paris, November 1989. 
Fischler M.A., Tenenbaum J.M., Wolf H.C. Detection 
of roads and linear structures in low-resolution aerial 
imagery using a multisource knowledge integration 
345 
technique, Computer Graphics and 
Processing 15 : 201-223, 1981. 
Image 
Groch W.D. Extraction of line shaped objects from 
aerial images using a special operator to analyse the 
profiles of functions Computer Graphics and Image 
Processing 18 : 347-358, 1982. 
Haralick R.M., King-Sun Fu Pattern recognition and 
classification, Manual of remote sensing, American 
society of photogrammetry, pp.793-804. 
McKeown M., Harvey W.A., McDermott J. Rulebased 
Interpretation of aerial imagery IEEE Workshop on 
principles of knowledge based systems, 1983. 
Renouard L. Experiences with automated terrain 
extraction from SPOT data, Proceedings of the 10th 
Earsel symposium, Toulouse, June 1990. 
Yee B. An expert system for planimetric feature 
extraction IGARSS 87 symposium, Ann Arbor, May 
1987. 
Zhang Zi-Jue, Shimoda H., Fukue K., Sakata T. A 
new spatial classification algorithm for high ground 
resolution images, IGARSS 88 symposium, 
September 1988, pp. 509-512.