ch (CRERA), 1 (2),
+ Hierarchies and
disation so difficult”.
pp22-28.
+ Hierarchies and
Standardisation so
vol. 6, No 3.
Into The Theory Of
Modelling In Geo-
Department of Land-
eningen Agricultural
"Delta-X Federated
iem". Proceedings of
on System for data
n, Ottawa, Canada.
(eds) Vol 30 No. 2.
nsing sector, natural
icia-solaco M., 1993
es in Interoperable
IP WG2.6 Database
le Database Systems
20 November, David
n Sacks-Davis (eds),
. Metadata Approach
ceedings of the 17th
arge Databases, pp.
Detection and extraction of complex map symbols
Ruedi Boesch
Federal Institute for Forest, Snow and Landscape Research (WSL)
CH-8903 Birmensdorf, Switzerland
e-mail: ruedi.boesch ? wsl.ch
Working Group III/2
Key words: cartography, discrimination, extraction, shape descriptors, triangulation, vectorisation
Abstract
The problem of extracting distinct map symbols from raster maps will be addressed in this paper. To enable natural scientists to
analyse the nation-wide distribution of natural objects such as avalanche obstructions, dry channels, tree groups, a semi-automatic
method has been developed to detect specific symbols from scanned topographic maps. The Swiss Federal Instituteof Forest Snow
And Landscape Research (WSL) uses the published map scale 1:25‘000 of the Swiss Federal Office of Topography (L+T).
After labelling and tracing the binarised raster data, shape descriptors like area, perimeter, moments, elongation, eccentricity, skele-
tons, Euler number and Fourier descriptors are calculated and stored in an image symbol database. In an interactive process, the
user defines the best fitting discrimination parameters based on the shape descriptor values. A local Hough transformation im-
proves the detection rate for line symbols such as found for avalanche obstacles.
Shape descriptors allow to identify map symbols like single trees, observation towers and triangulation. To detect complex map
symbols such as dry channels or avalanche obstacles, a distance-weighted triangulation is used to build a tetrahedron-like data
structure called tetra-tree. The tetra-tree allows to analyse and classify the spatial distribution of the primitives found with shape
descriptors. Generalised orientation and the convex hull of complex map symbols can be calculated directly from tetra-trees.
Some implementation details and generic limitations will be discussed.
1. Introduction
Landscape ecologists, biologists and geographers need data
about the existence, frequency and spatial distribution of
specific objects contained on maps or aerial images. Such ob-
jects include single trees, bushes, hedges, forest edges, dry
channels, tree nurseries, orchards or avalanche obstructions
(Figures 1a-1c). The map symbols for these can be defined as
aggregations of simple symbols such as points, lines, circles
and rectangles and are termed complex map symbols.
Major efforts have been made in the past to detect map sym-
bols [Báhr 1995, Lin 1994, Stengele 1995, Weber 1988]. In
most projects the methods of recognition are based on the
scanned map as a whole. Different types of lines and their re-
lated topology, houses and text labels are of major interest to
cartographers.
Figure 1a: avalanche obstacles -
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
Natural objects such as forest, bushes or terrain-related map
symbols are often neglected or have less priority. Fully auto-
matic map vectorisation remains still to be achieved [Lütjen
1987, Meyer 1993] and general pattern recognition on maps
will therefore remain a major research topic for the next years.
Currently, methods for the acquisition of complex map sym-
bols are mainly basedon interactive definition or manual digi-
talisation.
Figure 1b: dry channels
