International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
images to maps or to texts. The vertical path follows the
traditional scheme and is self-explaining. However, it is a new
perspective to consider steps from images to maps and/or to
texts as abstraction. The justification is that transformations
from verbal to pictorial and vice versa can never keep the
complete data set from each of the two representations.
Finally, it should be pointed out that semantics is not included
in Fig.1. Here, the features extracted from the "real world" are
restricted to geometry, topology, names, and words. In
computer science the term “semantics” suffers very often from
degradation to mere “attribute”, although it stands for
“meaning”, a very challenging concept. Integration of meaning
in analysis of pictures and texts absolutely requires context. An
isolated pixel, a separated line, or a single word carries no
meaning at all until it is put into its context.
3.2 Complementary terminology for verbal and pictorial
representation of geospatial descriptions
Since the human brain associates empirically the different
representations of knowledge — verbal or pictorial -, he is
normally not aware that he uses the same concepts for both
domains.
The term “understanding” is a metaphor from language applied
in image analysis. In the latter domain, it may replace “complete
image description”
“completeness” refers to sufficiency with regard to a particular
performance: a manual, describing the operation of an espresso
machine, or an image, displaying the complete road net of a
particular region.
“neighbourhood” relates to geometric or semantic vicinity.
Concepts, verbal descriptions, or styles may be called similar as
well as proximity of objects in imagery or adjacent figures in
graphs.
“readable” is called a text which is mentally accessible and does
not raise any doubts. It is, again, used metaphorically for maps
and imagery, where it means easy interpretability due to
adequate scale, layout, and design.
“precision” is a quality measure defined quantitatively for
geometrical features in maps, graphs, and pictures but also in
texts where it stands for a high level of detail in descriptions.
“homogeneity” indicates for both pictorial and verbal
representations low variance within a region or a textual
passage.
“level of abstraction” is used in a similar way for generalised
texts and maps.
“redundancy” means repetition of information or unnecessary
volume in verbal or in graphical descriptions. Redundancy is
not necessarily negative as it may stabilise a system, e.g. for
stochastic observations in least squares adjustment.
Nevertheless, in formal or deterministic systems, like coded
language or graphs, it is to be avoided.
“context” is an essential term for both the verbal and the
pictorial world which has already been mentioned in the
previous chapter. Different from the machine, the human puts
any signal “automatically” into a context and thus adds
semantics (=meaning), regardless of correct or incorrect
reasoning.
The complementary nature of graphics and language is not
limited to same or similar use of terms. It is true for the
processing level, too, and allows parallel analysis and synergetic
fusion of both domains. This will be shown in more detail by
the applications given in the chapters 4 and 5. As a first
example, “transformation” in relation of texts and images will
be analysed:
“Transformation” is defined as rule-based changing texts or
images from the original (input) level to a processed (output)
level. In case of leaving the facts (e.g. of messages) or the
objects (e.g. in images) unchanged, this transformation merely
gives “a different perspective of view". A text may be
formulated in many ways without changing the facts; an object
may be imaged in many ways without changing the object itself.
The different perspectives of view do not distort neither facts
nor objects.
4. EXAMPLE I: BRAZILIAN CADASTRE
4.1 Initial situation
The Brazilian cadastre — unlike most systems used in Europe or
America — is not map-based but consists of millions of
notarially certified texts, each describing the boundaries of
single land parcels. Additional maps are rarely available, and in
cases of juridical conflicts, only the texts are committing. This
form of administration of land ownership does not meet the
necessary requirements to guarantee the legitimacy of
ownership or to allow an efficient update and management of
ownership information.
4.2 Data
Data base for this project are approximately 70 examples of real
texts and accompanying maps. An example map can be seen in
Fig. 2.
x ns pred - AMA s
+ ownerC . on X“ | owner B
* unfer former owner Z
«08 ; ci
3 ao ^ -
A e i 39 gt me =
‚eo
t or
N sat
: e : : =
«o9 Sw? ?
\ e
+ e
owner A E
owner A
former owner Z
Figure 2. Part of a map of the Brazilian cadastre (real names of
the property and owners replaced due to protection of data
privacy). The circle indicates the starting point of the textual
description (not marked in the original map).
The associated text to the map of Fig. 2 (translated from
Portuguese) is presented below:
“It starts at the point of conjunction between a property of
218
Interi
owne
belon
straig
and |
68.70
belon
creek
m, w
to Z)
351.6
it tur
slight
up to
on th
AS Ct
parts
adjac
infori
value
4.35
The
"easy
outpu
(line-
proje
map |
of ab
comp
and tl
comp
repre:
More
symb
parce
repre:
land
possil
point:
origir
A clo
at fir:
comp
comp
referr
the te
lengtl
existi
Howe
infor
and v
44 "
In orc
level,
chose
they €
withit
applic
image
2000;
the a
(Nien
an ex
