Bariloche
National
Park
Hyde Park, --^ 5
Corner
Part-link
Specialization- link
— ——- Instance link
Meeting
Places
Figure 2: À semantic net
5 CONFUSION DUE TO UNPRECISE CONCEPTS
There are concepts and terms which are overloaded with se-
mantic content and therefore have become unprecise. Lin-
guists call them homonyms and we use them in everyday life
without being aware of this phenomenon. If we take, for ex-
ample, the term "garden", we see that it means: a place
where flowers are grown in "flower garden" ; it does not mean
a place where children are grown in "kindergarden" but a
place where very small children receive their first education.
A "japanese garden" is not a garden in Japan but a certain
arrangement of small ponds and bridges, trees and bushes,
grass and flowers not to speak of "the Garden of Eden" the
mental image of which surely differs in all of us.
Only when we are asked to explain concepts do we discover
how difficult it is to come to a clear definition, to choose
the distinctive characteristics, those which describe the most
important features or the features of interest. For example,
how can "sealed surfaces" be described? Which are the key
elements? Constructors involved in the construction of build-
ings might understand " roofs, walls and floors covered by a
waterproof material so as to keep humidity out of rooms";
urbanists would define sealed surfaces by "surfaces that do
not absorb rainwater”. Photogrammetrists on the other hand
might define them as surfaces having a certain spectral signa-
ture. Words like " park" as represented in Figure 2, may stand
for a special type of public greens, an existing area within an
urban settlement or a surface foreseen in a master plan to be
developed into a park in the future; moreover, the concept
also covers a specially beautiful sector of the landscape which
is protected by law.
10
A single term with multiple meanings can be compared to a
single spectral signature which represents different land use
classes in Remote Sensing (Fig 3).
Figure 3 shows land use classes in a confusion matrix. Forest,
meadow, swamp and park might show similar sectral signa-
tures and might accordingly be assigned to the same class.
An object "square" could be assigned indistinctly to "sealed
surface”, " road" , "parking site", "industrial area" and "resi-
dential area". The question which is the correct designation
in each case is clarified through context.
This kind of non-identity of the representations is an enor-
mous problem for semantic modelling.
6 CONFUSION DUE TO INACCURATE
DISTRIBUTION
Another source of inaccuracy and uncertainties when it comes
to the description of objects at the symbolic level are unsharp
boundaries. In the real world sharp boundaries do not exist
but are individually defined by discrete assignment.
This also applies to apparently simple concepts like " build-
ing". In the German cadaster, for example, not only the
building, the constructed structure but the whole parcel is
given the label "bebaut" (occupied by a building) no matter
whether such structure occupies the whole lot or not.
Instances for objects with imprecise definitions are all forms of
"mixtures" (e.g. mixed forest or biotope). The distribution
of species and the occurrence of disease also fall into this
category.
The problem of unsharp boundaries has to do with inexact
quantities. When does "water" deserve the label "polluted
water”? Where is the limit between "scarcely populated"
area and "densely populated” area? Still worse: where are
the boundaries of objects which due to their very nature do
not have clear cut limits (mountains and valleys) and are
therefore defined without precision?
Errors resulting from the combination of this type of ele-
ments are extremely difficult to model, especially when the
term designating such phenomena is polysemous (has many
meanings). The stochastic description of geometrical param-
eters appears to be almost trivial when compared to the cor-
responding modelling of semantic parameters.
7 CONFUSION DUE TO INAPPROPRIATE
METHODS
Linguistic difficulties with semantic modelling are not limited
to terminology, to concepts and terms. They are overlapped
by a phenomenon related to methodology and to graphic
primitives.
An example will make this clear: we use the tools of Analyti-
cal Geometry for the semantic modelling of images although
neither points nor lines exist in the real world or in photo-
graphical images, not to mention "digital" ones. Neverthe-
less we use such elements and the related models without
being aware of the difference with reality. It is not surpris-
ing, therefore, if such models are only able to model reality
in quite an imperfect way. The convention " points and lines"
belongs to the domain of the symbolic level. They are arti-
ficial concepts, genially conceived and have been useful for
thousands of years but they are foreign to the real world.
Another example of the fact that semantic modelling still
clings to concepts and metaphors which are characteristic
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B6. Vienna 1996
of 1
con
"C
me:
nat
of "
rela
Thi
por
cap
imp
cap
helr
con
tion
alw
cap
"Re
Ori,
pon
me:
mat
For
resc
tor
tod
its
resc
obj:
The
tior
ima
tecl
the
can
ture
