grammetric research. This framework needs to be em
bedded into the activities of the surveying and mapping
community. They can be described as an evolving cycle
(fig. 1), where the object, namely our environment, in
cluding its topography, is observed, documented in maps
used for planning, and changed to serve the ecological
balance, or other needs. Photogrammetry and remote
sensing here appear to be the technology for the auto
matic analysis 1) of image data leading to an interpre
tation, thus uncovering the information about the objects
for the use in (digital) maps.
Image interpretation
Image interpretation up to now still is mainly performed by
human specialists. It is not well understood, though per
formed with great success. In order to be able (at least
partially) to transfer this human capability to a machine a
detailed modelling is required.
Machines only can handle symbols using some kind
of specified formalism. Semiotics is the theory about
handling symbols (fig. 2). It deals with the relations of
symbols to other symbols (syntax), to objects (semantics)
and to the use of symbols (pragmatics).
other symbols
symbol
real world object
syntax
semantics
pragmatics
user
Fig. 2. Semiotics according to Bochenski [6], Formalisms map se
mantics to syntax, requiring semantic integrity, i.e. proper
meaning of all symbols and rules.
In our context the pragmatics reflects the different re
quirements of the users of photogrammetry and remote
sensing and the different types of models to be used
when analyzing image data. The same symbol, e.g. a
bright spot, may have different context dependent inter
pretations: it may be treated as an outlier, as a building
or as a factory depending on the model used for image
analysis. The need to specify the purpose or goal of an
interpretation task explains the urgent need for meta
data, i.e. explicit description of the data. The necessity to
describe context or relevance to a specific application is
equivalent to specifying the pragmatics of an interpre
tation scheme.
Given a certain context specifying the semantics is identi
cal with modelling the objects and their appearance2).
This includes the description of all necessary relations
between different objects, between objects and their
parts and their change over time. This modelling has
been realized to be the bottle neck for developing
automatic interpretation systems. Existing interpretation
systems contain more or less explicit object and image
2) This goes beyond the modelling in Geo-information systems
where the appearance of objects is not required.
models, they, however, appear shallow compared to the
models used by human analysts.
The mutual relations between the used symbols, i.e. the
syntax, for describing objects, relations or images may be
poured into grammatical or other rules. Such grammars
may describe formalisms or (artificial) languages which
we know from mathematics (theories) or programming
(algorithms). The notion semantics is used in two closely
related forms (fig. 3):
the semantics of a symbol, a symbolic structure, a
concept or a notion is their relation to a real object or
structure. One also could say: the interpretation of a
symbolic structure defines its semantics. This is
semantics as part of semiotics [6];
relations between concepts or notions, only existing in
the model, are termed semantic relations as they
restrict the semantic relation of the concepts or
notions, which themselves are symbolic structures, to
real objects or structures. This is semantics within
knowledge representation [75],
We will use both aspects of semantics throughout this
paper.
Within the semiotic framework using formalisms, realized
as computer programs, needs a mapping of the seman
tics of the symbols to their syntax. This mapping requires
a careful design such that all rules within the syntactic
framework map to rules which hold in reality, in order to
avoid inconsistencies and to keep semantic integrity.
The formal description of the used rules itself uses a
language, i.e. a formalism with different symbols and its
own rules. For example, in data bases this is the data
description language. Therefore we at least have to dis
tinguish three levels: the level of the real objects, which
are no symbols, the level of the symbols (words) standing
for the objects and their relations (object language)
and the level of the formalism in which we describe the
structure of the object language (meta-language).
Symbols, notions or words of the meta-language are not
to be confused with symbols, notions or words of the
object-language in order to avoid contradictions, conflicts
or paradoxes. As the meta-language again has to be
described we arrive at a multi-level description of the
models we are using. The possibility to formalize knowl
edge obviously decreases with increasing level of ab
straction finally leaving us with the natural language,
interpretation,
in ter pretation,
semantics
o bje c t
concept
concept
object
Fig. 3. Relations between concepts (notions) in a model and real
world objects: relations between concepts are semantic
relations, interpretations of concepts, i.e. their relations to
real objects define the semantics of a concept.
Warning: In object oriented programming, but not only the
re, often concepts are treated as objects. Then objects in the
model and objects in the real world have to be distinguished.
NGT GEODESIA 93 - 8
373
