Steffen Frischknecht, Entela Kanani
Institute of Geodesy and Photogrammetry
Swiss Federal Institute of Technology (ETH) Zurich
ETH Hoenggerberg, CH-8093 Zurich, Switzerland
Tel. ++41-1-633 3045 or 3055, Fax ++41-1-633 1101
Email: stf@geod.ethz.ch or kanani@geod.ethz.ch
Abstract. This paper presents some results of an ongoing project
in the field of pat-tern recognition and automatic vectorization. The goal
of the project is to obtain automatically multiple information and structured
objects from digital topographic maps in raster format. An approach of
knowledge-based Template Matching with very high recognition rates (about
95%) and a procedure to segment and isolate coherent areas of a colour
layer are presented. All segmented areas are encoded uniquely, which allows
to calculate features for every area. The segmentation and the features
can serve as a preselection in the raster image to obtain information about
regions or objects of high, low or no interest. Furthermore, a new procedure
that provides an automatic vectorization of areal objects is shown. The
vectorization occurs by adjusting the end points of straight line segments,
facilitated by robust estimation functions which are resistant to falsely
assigned parts of areal objects and stable with respect to deviation from
the given distributional model. The process starts with a simple model
(rectangle) and gradually uses more complicated ones (right-angle polygon,
simple polygon, circle etc.). If a more complicated model does not give
a better representation, the simplest model is adopted. The resulting data
can be exported as DXF to standard CAD-packages or GIS-Systems for further
use.
The methods presented have been tested in many topographic maps and found
to be operationally and qualitatively suitable for automatic processing.
Chair of GIS at Institute
of Geodesy and Photogrammetry
Institute of Geodesy and Photogrammetry
of ETH Zurich
Homepage of ETH Zurich
Homepage of
Steffen Frischknecht
Homepage of Entela
Kanani