ition
such
ons.
the
tiple
also
such
ig in
‚Le.
nore
is is
and
ales
ules
ing.
tore
ere
rom
odel
1Ses
suit
the
and
Mul-
just
data
2 di-
) re-
data
tion.
rage
> Going up in the hierarchy, a reduction of unnecessary
details and at the same time an emphasis of the impor-
tant ones is achieved. The different levels of detail may
reveal essential information for different users.
> A severe problemin GIS is the fusion of data originating
from different sources. Data can be captured based on
different data models and also with different data qual-
ity. The data sets are similar in the sense that they are
captured at approximately the same scale. In the con-
text of the so-called conflation [Walter & Fritsch 1995]
map matching techniques are applied to determine cor-
responding parts between the data. These methods
also have to allow for partial matches. This problem
especially occurs when thematic data has to be inte-
grated into general topographic data sets.
> In image analysis multiscale representation is also an
important issue. In order to get approximate values
for interpretation or matching, usually coarse-to-fine-
approaches are selected, using image pyramids (e.g.
[Hahn 1989]) but also a series of symbolic descriptions
([Bobick & Bolles 1989]).
In GIS the following multiple representation problems can be
distinguished, which are visualized with topographic data of
Germany:
GDF —> DLM25
Thematic Data
DLM200
|
DLM1000
Map 1:1Mio
Y
Figure 1: Examples for multiple representation problems, vi-
sualized with topographic data available in Germany: ATKIS
DLM's (Autoritative Topographic-Cartographic Information
System) at three different resolutions, GDF (Geographic Data
File: standardized data exchange format in Europe for road
traffic data).
model generalization: a bottom-up aggregation (general-
ization) of objects going from one scale to the next (e.g.
transition from DLM25 to DLM200 (Digital Landscape
Model 1:25 000 and 1:200 000, resp.)),
cartographic generalization: generalization going from a
data base representation to a graphical representa-
tion (e.g. cartographic presentation of contents of
DLM1000 in a map of scale 1:1Mio),
conflation: matching of data sets of different origin, but de-
scribing the same physical reality (e.g. fusion of GDF
road data and ATKIS road information).
769
Whereas the cartographic generalization can be considered
as a problem of high complexity (especially the problem of
displacement), solutions for the model generalization seem
to be closer at hand. The question even rises, if cartographic
generalization will ever be achieved - or if it should rely on
semi-automatic processes (e.g. with commercial products
like MapGeneralizer from Intergraph). Thus the proposed ap-
proach refers to the model generalization only.
The term generalization is normally used for cartographic
generalization. There it implicates, that the generalized ver-
sion of an object completely replaces the original one, since
it is no longer needed after visualization. Data abstraction or
model generalization however results in a a hierarchy of ob-
jects - where all manifestations exist side by side with equal
rights. In the sequel generalization will however be used to
describe both types of data abstraction.
1.2 Possible Realizations
A spatial database comprising multiple levels of detail can be
organized in various ways. It can be realized by having a sin-
gle most detailed representation in conjunction with tools to
derive a series of other layers of different scales. The other
possibility is to keep multiple representations of the objects
on different pre-given levels of detail in the system. The ad-
vantage of the first alternative is that only one data set has to
be stored, which can be managed and accessed consistently.
In the second case redundant data has to be dealt with. On
the other hand the time for the calculation of data at a certain
scale has to be taken into account. Also - to date - no efficient
generalization algorithms are available. Furthermore there is
no tool to propagate updates through a series of derived data
sets - which is an important issue for database revision.
Ideally a GIS should comprise all possible information - ev-
ery application then should be able to deduce the problem
specific information from it. This presumes to have rules for
the transition of representations of objects between different
levels of detail. Another still unsolved problem is the selec-
tion of the optimal scale for a given task. Thus the following
questions arise:
> Which objects have to be represented in a certain
level ?
> How are these objects represented (point, symbol, line,
area, ...)?
> When do the objects disappear and how ?
In general, the national mapping agencies have already an-
swered the first question. E.g. the surveying authorities in
Germany have detailed descriptions of the objects to be rep-
resented in a map of a certain scale (ATKIS ([Harbeck 1995])
- a general framework for a digital representation of the map
data ). These descriptions, so-called object catalogues, also
include the way the objects should be captured, their ac-
curacy and representation in terms of geometric primitives
(point, line, polygon). They are however only given for cer-
tain discrete scales (scales 1:25 000, 1:200 000, 1:1 Mio). In
order to generalize to other intermediate scales, such cata-
logues have to be established accordingly. There is a great
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996