OT XS data (Fig. 4)
els from all line pixels
e pixels of Fig. 6 by
t has been vectorized
nes and the methods
e represents the river.
almost completely. Additionally, there are small fragments of
roads inside cities.
The use of multispectral images is not suitable for distinguish-
ing all kinds of lines in an image, e.g. it will be impossible to
recognize railways from SPOT or TM data in this way. It is
necessary to make use of the knowledge that we find in exist-
ing maps or Geographic Information Systems and to combine
it with the image using matching techniques.
5. CONCLUSIONS AND OUTLOOK
The results of this paper show that the presented models
allow to automatically extract lines and edges from digital
images. The extraction of lines as well as edges leads to the
segmentation of the lines in the image. Even in cases where
the width of lines is close to the limit of the resolution of
the image, it was possible to find lines. Due to the spatial
resolution of the operational remote sensing sensors SPOT
and TM feature extraction from these images typically leads
to a lot of spurious details. We face this fact by two different
means. One is the robust estimation of a threshold that
enables us to select features rising above the noise level. The
other is the use of spectral characteristics of linear features
which showed that it is possible to distinguish roads and rivers
among a large variety of other lines, which is demonstrated
by the example.
Further work on the presented methods aims at incorporating
more knowledge. For instance, the use of knowledge about
autobahn junctions or knowledge about the radii of curves of
railways and roads is promising. Sophisticated line trackers
(Grün and Li 1994) will improve gap closing and make the
detection and analysis of nodes more robust. This enhance-
ment of the results especially at intersections or junctions is
important since they are the frame for matching the extracted
objects with the objects in an existing Geographic Information
System.
In this paper the methods have been applied to satellite im-
ages only. But it is possible to make use of them for road
extraction from aerial images (Li et al. 1992, Ruskoné et al.
1994a, 1994b), too. The link to those methods is an image
pyramid which starts at coarse resolution improving the re-
sults in subsequent steps by making use of the approximations
achieved earlier.
REFERENCES
Busch, A., 1993. Ein Verfahren zur Erkennung linien-
hafter Objekte in digitalen Bildern. In: Póppl, S.J., Han-
dels, H. (Eds.). Mustererkennung 1993. Springer-Verlag,
Berlin, pp. 616-623.
Busch, A., 1994. Fast recognition of lines in digital images
without user-supplied parameters. In: Ebner, H., Heipke, C.,
Eder, K. (Eds.), International Archives of Photogrammetry
and Remote Sensing, Vol.30, Part 3/1, SPIE — The Interna-
tional Society for Optical Engineering, Washington, pp. 91-
97.
Canny, J., 1986. A computational approach to edge detec-
tion. IEEE Transactions on Pattern Analysis and Machine
Intelligence 8, pp. 679-698.
Deriche, J., 1990. Fast algorithms for low-level vision.
IEEE Transactions on Pattern Analysis and Machine Intel-
ligence 12, pp. 78-87.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
Grün, A., Li, H., 1994. Semi-automatic road extraction by
dynamic programming. In: Ebner, H., Heipke, C., Eder, K.
(Eds.), International Archives of Photogrammetry and Re-
mote Sensing, Vol.30, Part 3/1, SPIE - The International
Society for Optical Engineering, Washington, pp. 324—332.
Haralick, R.M., Watson, L.T., Laffey, T.J., 1983. The topo-
graphic primal sketch. Int. J. Robotics Research 2, pp. 50-72.
Reprint in: Kasturi, R., Jain, R.C. (Eds.) 1991. Computer Vi-
sion: Principles. IEEE Computer Society Press, Los Alamitos,
pp. 195-217.
Haralick, R.M., 1984. Digital step edges from zero crossing of
second directional derivatives. IEEE Transactions on Pattern
Analysis and Machine Intelligence 6, pp. 58-68.
Haralick, R.M., Shapiro, L.G., 1992/93. Computer and Robot
Vision, Vol. 1 and 2. Addison-Wesley, Reading.
Koch, K.R., 1985. Ein statistisches Auswerteverfahren für De-
formationsmessungen. Allgemeine Vermessungs-Nachrichten
92, pp. 97-108.
Koch, K.R., 1990. Bayesian Inference with Geodetic Applica-
tions. Springer-Verlag, Berlin.
Li, S.Z., Kittler, J., Petrou, M., 1992. Matching and recog-
nition of road networks from aerial images. In: Sandini, G.
(Ed.). Computer Vision — ECCV'92, Second European Con-
ference on Computer Vision, Santa Margherita Ligure, Italy,
Proceedings. Springer-Verlag, Berlin, pp. 857-861.
McKeown, D.M., 1994. Top ten lessons learned in automated
cartography. Paper presented at ISPRS Comm. Ill Sympo-
sium, Munich, Germany, Sept. 5.—9. 1994.
Ruskoné, R., Airault, S., Jamet, O., 1994. Road network
interpretation: A topological hypothesis driven system. In:
Ebner, H., Heipke, C., Eder, K. (Eds.), International Archives
of Photogrammetry and Remote Sensing, Vol.30, Part 3/2,
SPIE — The International Society for Optical Engineering,
Washington, pp. 711-717.
Ruskoné, R., Airault, S., Jamet, O., 1994. A road extrac-
tion system using the connectivity properties of the net-
work. Zeitschrift fiir Photogrammetrie und Fernerkundung
62, pp. 174-180.
Sahoo, P.K., Soltani, S., Wong, A.K.C., 1988. A survey of
thresholding techniques. Computer Vision, Graphics, and Im-
age Processing 41, pp. 233-260.
Schickler, W., 1992. Merkmalsextraktion fiir MeBaufgaben in
der digitalen Photogrammetrie. Ein statistisches Auswertev-
erfahren für Deformationsmessungen. Zeitschrift für Pho-
togrammetrie und Fernerkundung 60, pp. 107-116.
Schulz, B.S., Wende, C., 1994. Vollautomatische, hochdif-
ferenzierte, pixelweise Klassifizierung multispektraler Bilder
— eine neue Methodik —. In: List, F.K. (Ed.), Vorträge
13. Wissenschaftlich- Technische Jahrestagung der DGPF,
Geoinformation durch Fernerkundung. Publikationen der
Deutschen Gesellschaft für Photogrammetrie und Fernerkun-
dung, Band 2, Berlin, pp. 193-197.
Venkatesh, S., Rosin, P.L., 1995. Dynamic threshold determi-
nation by local and global edge evaluation. Graphical Models
and Image Processing 57, pp. 146-160.
93
