GRAPHICS AND LANGUAGE AS COMPLEMENTARY FORMAL REPRESENTATIONS 
FOR GEOSPATIAL DESCRIPTIONS 
Hans-Peter Bähr and Marina Müller 
Institute of Photogrammetry and Remote Sensing, University of Karlsruhe, Englerstr. 7, 76128 Karlsruhe, Germany — 
(baehr, mueller)@ipf.uni-karlsruhe.de 
KEY WORDS: Abstraction, Spatial Analysis, Knowledge Representation, Navigation, Mapping, Disaster Management 
ABSTRACT: 
In image understanding, the importance of language (e.g. the choice of class names in image analysis) has been widely 
underestimated up to now. Within this paper, the close interaction between language and graphics is analysed for geospatial 
applications. It is shown that texts and maps are two complementary representations of knowledge, resulting among other in the use 
of an equal terminology for verbal and pictorial representations. Two different real scenarios (Brazilian cadastre and message 
analysing systems for disaster management) are introduced where the direct relationship of language and image analysis becomes not 
only obvious, but the mutual transformation - of one representation into the other - is the essential task of the application. First results 
of the presented project already demonstrate how both forms of representation can be transformed to a common symbolic level as a 
basis for further synergetic analyses. 
KURZFASSUNG: 
Bei Bildverarbeitungsaufgaben wurde bisher die Bedeutung von Sprache (z.B. bei der Wahl von Klassennamen in der Bildanalyse) 
weitgehend unterschätzt. In dieser Veröffentlichung wird die intensive Wechselbeziehung zwischen Sprache und Graphik 
eingehender analysiert. Texte und Karten stellen zwei komplementäre Formen der Wissensrepräsentation dar, wie es u.a. beim 
Gebrauch der gleichen Terminologie für verbale und bildhafte Repräsentationen deutlich wird. Anhand von zwei 
Anwendungsszenarios (brasilianisches Kataster und Systeme zur Analyse von Meldungen innerhalb des Katastrophenmanagements) 
werden nicht nur die direkten Beziehungen zwischen Sprache und Graphik veranschaulicht, sondern es sind auch konkrete Beispiele, 
bei denen die Transformation von einer Reprüsentationsform in die andere die wesentliche Aufgabe darstellt. Erste Ergebnisse des 
vorgestellten Projekts zeigen, wie Graphik und Sprache auf eine gemeinsame symbolische Ebene transformiert werden kónnen, auf 
deren Basis integrierte Analysen erst móglich werden. 
1. INTRODUCTION 2. LINGUISTIC ELEMENTS IN GEOSPATIAL 
ENVIRONMENT 
Language has always been considered as a prominent tool for 
knowledge acquisition, analysis, and storage. Language in “Geospatial environment” in the context of this paper integrates 
spoken or written form is the dominant representation of any georeferenced descriptions. It comprises their analogue and 
knowledge in all its very different facets. their digital representation: images, maps, graphics, and other 
kind of 2D data, so to say “clippings” from the “real world”. 
However, it will be shown that terms, concepts, language 
together with meaning and semantics interestingly plays only a Navigation marks an application where language has already 
minor role in image analysis applications. On the one hand, been introduced (Kray and Baus, 2003). This is, for instance, 
language defines an attractive field of research of its own e.g. in true for navigation of pedestrians, but especially for navigation 
automatic understanding of spoken language, automatic reading of vehicles, which are tracked by GPS. The GPS output is 
and classification of handwriting, or recognising persons by restricted to coordinates, which do not give any meaning to a 
their individual way of speaking. On the other hand, little has human unless referred, e.g., to a street (i.e. a name). Piloting a 
been done up to now to define language — at least names or car driver requires spoken language as the optimal tool. Map- 
single words — as a data type within image analysis. based navigation leads to unnecessary and sometimes dangerous 
It is exactly this apparent gap which will be focused in the distraction of the car driver. 
following text. 
In the course of classifying geospatial data sets, like satellite 
The situation has been given attention by the Institute of imagery, physical parameters have very rigorously been 
Photogrammetry and Remote Sensing at Karlsruhe University modelled, not to speak of the classification algorithms. Classes 
since the early 90ties, but only recently a research project are completely described by spectral signatures or even by 
“Abstraktion graphisch und verbal repräsentierter geometrical or topological features. However, much less 
Geoinformation” was established, embedded in a major bundle- attention is given to the major parameter: the meaning of that 
type research programme! sponsored by the German Science class, represented by its name. This situation had been pointed 
Foundation (DFG). out by (Bähr and Schwender, 1996). The name of a class 
including their semantics denotes the final goal of the 
4 classification process which is necessarily incompletely 
This programme is entitled “Abstraktion von Geoinformation described by the number of physical, spectral, and geometrical 
bei der multiskaligen Erfassung, Verwaltung, Analyse und models. 
Visualisierung” chaired by Monika Sester/U of Hannover 
  
216 
Intern 
Nota | 
param 
forest. 
rigoro 
and of 
Finall 
langu: 
case c 
or wri 
the a 
geosp: 
and a 
develc 
applic 
2003; 
in this 
The n 
data, 
unders 
When 
spatial 
shared 
image 
and L 
define 
3.1 TI 
Both | 
differe 
definit 
Logic’ 
Minsk