RECOGNITION OF HATCHED CARTOGRAPHIC PATTERNS
Regine Brügelmann
Institute of Photogrammetry and Remote Sensing
University Karlsruhe
Germany
bru@ipf.bau-verm.uni-karlsruhe.de
Commission Ill, Working Group 3
KEY WORDS: Cartography, Digital, Map Interpretation, Image Understanding, Pattern Recognition, Automation
ABSTRACT
This paper deals with the automated interpretation of large-scaled scanned maps, using the example of the german base map
Deutsche Grundkarte 1:5000 (DGK5). The goal is a raster-to-vector conversion of the map content. The increasing demand
of digital data for building databases in Geographic Information Systems requires the development of powerful techniques to
support automated map understanding. The paper presents an approach which automatically detects buildings and separates
them from the remaining map objects. In the (mostly) black and white DGK5 map buildings are represented by their outlines
filled with hatched patterns. In contrast to most of the existing approaches for cartographic pattern recognition, a raster based
method is proposed. The typical sequences of black and white pixels forming the hatched patterns are used to detect the
buildings. The used raster based methods involve the investigation of runlength encoded image rows and columns, a kind of
directional region growing and operations of mathematical morphology. It is shown that this map understanding task can be
solved in the raster environment up to an advanced processing stage.
KURZFASSUNG
Dieser Artikel befaßt sich mit der automatischen Interpretation von großmaßstäbigen gescannten Karten am Beispiel der
Deutschen Grundkarte 1:5000 (DGK5). Ziel ist eine Raster-Vektor-Konvertierung des Karteninhalts. Der steigende Bedarf an
digitalen Daten für den Aufbau von Datenbasen in Geoinformationssystemen erfordert die Entwicklung leistungsstarker Meth-
oden für das automatische Kartenverstehen. Dieser Aufsatz beschreibt einen Algorithmus, der automatisch Gebäude lokalisiert
und sie von den übrigen Kartenobjekten separiert. Gebäude werden in dieser vorwiegend schwarz-weißen Strichkarte durch ihre
Umrisse dargestellt, die mit Schraffur gefüllt sind. Im Gegensatz zu den meisten bestehenden Arbeiten im Bereich der kar-
tographischen Mustererkennung wird hier ein rasterbasierter Ansatz vorgeschlagen. Die typischen Abfolgen der schwarzen und
weißen Pixel, die die Schraffur bilden, werden für die Detektion der Gebäude genutzt. Die dabei benutzten rasterbasierten Tech-
niken umfassen die Untersuchung von lauflängenkodierten Bildzeilen bzw. -spalten, ein richtungsabhängiges Regionenwachstum
und Operationen der Mathematischen Morphologie. Es wird gezeigt, daß die gestellte Mustererkennungsaufgabe bis zu einem
fortgeschrittenen Stadium im Raster lösbar ist.
1 INTRODUCTION
1.1 Motivation
The demand of digital information is rapidly increasing due
to advanced computer technology and the widespread use of
Geographic Information Systems. Each GIS application re-
quests a georeferenced database. Existing paper maps repre-
sent such powerful databases. Before they can be integrated
in a GIS they have to be converted into a digital vector rep-
resentation. This is a time-consuming process if it is done
by manual digitizing. Scanning maps is a faster way of ob-
taining digital data. Unfortunately the primary output of the
scanning are raster data without any semantic information.
Thus raster-to-vector conversion is needed as a first step.
Image understanding algorithms can be used to facilitate
and accelerate the raster-to-vector conversion of maps. This
task belongs to the broad field of document image recog-
nition which already yields good results for optical charac-
ter recognition (OCR). Technical drawings such as engineer-
ing drawings, graphics and maps, of course, are much more
complex than pure alpha-numerical text information. Al-
though some work has already been done in the field of
automated map interpretation, e.g. (lllert, 1990), (Kasturi
et al., 1990), (Suzuki/Yamada, 1990), (Crosilla/Piccinini,
1991), (Antoine/Collin/Tombre, 1992), (Boatto et al., 1992),
(Hori/Okazaki, 1992), (Ablameyko et al., 1993), (Ebi, 1993),
82
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
(Yamada/ Yamamoto, 1993), (Mayer, 1994), operational sys-
tems interpreting complex maps in satisfying quality are still
rare.
The presented work deals with map interpretation using im-
age understanding algorithms aiming at automatic raster-to-
vector conversion. Because of the high complexity of map
graphics, the automated interpretation of the map as a whole
document is not possible at the current stage of technology.
Thus the presented work is focused on buildings. Spatial in-
formation about buildings are needed in many fields such as
regional planning or 3D modelling of urban areas. Further-
more, digital vector data representing the shape of buildings
can be used as apriori-knowledge to support automated anal-
ysis of actual aerial photographs for change detection and
map updating as it is, for instance, shown by (Quint/Bähr,
1994), (Quint/Sties, 1995).
1.2 Data Source
As data source the Deutsche Grundkarte 1:5000 (DGK5) is
used which is the primary topographic map of Germany cov-
ering at least 8096 of the country. The DGK5 represents the
topography as brown contours and all other objects as black
lines and symbols. The map objects are determined by their
shape, linewidth and relativ position. Fig.la shows a subset
(760 by 75 metres in reality) of a scanned DGK5.
This type o
to its large
The mappe
accuracy ra
data are mo
m? is mapp
which mere
the DGK5 ‘
In the DGK
a hatched
cases some
pattern con
types of bu
terns varyit
to the long
marked by
outbuilding
lines (angle
Figure
However, à
cause the c
single rect:
centres. |n
direction a
regularities
broken hat
graphic ot
hatching li
interruptec
non-paralle
2 O
Existing a
(ert, 19
1993), (M
the entire
try to finc
tion at thi
original inf
thinning, ¢
result, err
Another d
obtained \
various prt
In contras
building r
advanced
The devel
At the beg
features t
maining n
