Istanbul 2004
of Figure 8.
aggregated
n plots are
class of
; have been
opographic
on level 4
>d to
IS
lave its own
through a
This means
^ definitions '
e its own
te structures
jects at each
> example of
ind use. The
the fact that
ggregated to
level. There
een various
ies related to
between the
nse that the
|] objects, as
e transferred
n about the
Iso tôp-down
behaviour of
ained by the
level objects.
also been
stems theory
) they have
of landscape
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
5.3 Image analysis
Functional object aggregation hierarchies can also be used for
remote sensing data (image) analysis. We will follow the example
of (Zhan, 2003). Zhan used the combination of IKONOS Multi-
Spectral data and laser altimetry data obtained with the TopoSys 1
laser scanner. The example refers to an area of 3 x 3 km’ in the
south eastern part of Amsterdam. The TopoSys data have been
processed to generate height data with an accuracy of about 15 cm
in grid with 4 m spacing. The IKONOS data were interpreted into
cover classes per pixel, these were: built-up area, vegetation and
water. By combining these results with the height data they could
be interpreted in second step into the classes: building, road, open
paved area, grasslands, trees and water surface. These were class
labels assigned per pixel From these labelled pixels image
segments could be formed presenting spatial objects. The
identified buildings could be further classified based on their size
and the height data. These classification results were the input of
figure 10.a. Through a triangulation based technique spatial
clusters of similar objects were formed as in Figure 10.b. These
clustered objects were then aggregated into the urban land use
units of Figure 10.c. When the results of Figure 10.b are combined
with road data then the classified street blocks of figure 10.d are
obtained. We see that this method for image analysis has been
based on a semantic approach based on the functional object
aggregations, i.e. each aggregation step specifies functional units at
a higher level in the context of urban land use. In this example we
go from the level of classified pixels, to elementary objects and
then on to land use units or street blocks.
6. DISCUSSION AND CONCLUSIONS
The discussions in the previous sections of this paper illustrate
several issues with respect to the specification of the semantics of
spatial data. When discussing these issues we should keep in
mind that the focus of this paper was on object structured
approaches. The representation of spatial objects has two aspects:
I. The structure of such representations specifies which data
types play a role in these representations and what
relationships may occur between data of these different
types.
The semantics of such representations specifies how
database representations refer to "real world features”.
Semantics is understood here in the sense of (Winteraecken.
1987)
The previous sections dealt mainly with issues related to the first
aspect. The semantic issues were dealt with only implicitly in the
examples. From these examples we can learn that the semantics
of objects should always be understood in an application context.
Such a context generally refers to some spatial process be it
natural, man-induced, social or economic. Objects play then the
role of process response units so that their semantics can not be
specified without reference to such a process. Therefore the
semantics of object definitions will generally be hard to
understand outside the context given by such a process. This
implies that it will hardly be possible to standardise semantic
object definitions.
bo
The structural aspects of spatial object representations have been
formulated with in the context of formalized data models. These
models have a mathematical structure which helps us to
understand some fundamental constraints that should be fulfilled
to avoid ambiguity in spatial object representations. The relation
found in Section 3 between spatial and thematic partitions
appeared give an essential constraint for the specification of
objects and object classes.
CONCLUSION 1
The dual partition structure, i.c. the combination of thematic and
spatial partitions, is fundamental for the specification of
semantically unambiguous and complete spatial representations.
This conclusion implies that spatial objects can not be considered
in isolation because changing the geometry of one object,
removing an object or entering a new object has an effect on the
structure of spatial partitions. Similarly removing or entering
object classes has an effect on thematic partitions and possibly
also on geometric partitions.
CONCLUSION 2
The dual partition structure mentioned in Conclusion 1, implies
that spatial objects should not be seen in isolation. Spatial objects
should generally be understood as components of a spatial
complex.
Due to the dual partition structure temporal changes rarely affect
individual objects but affect complexes of objects, therefore:
CONCLUSION 3 :
Spatial objects should generally be understood as components of
dynamic spatial complexes.
We saw that due to the dual partition structure thematic class
generalization should be followed by a class (or similarity) driven
object aggregation. Functional object aggregation requires a newly
defined classification system for the aggregated objects. These
combined operations are required to maintain the dual partition
structure which is important to maintain the consistency of the
semantic specifications of spatial data sets.
CONCLUSION 4
The dual partition structure should be maintained when object
aggregation (generalization) procedures are applied to spatial
object data. This is important for maintaining the consistency of
the semantic specifications of spatial data sets at each aggregation
(generalization) level and for maintaining the consistency between
levels.
Earlier in this section we stated that the semantics of spatial object
representation should be understood in the context of spatial
processes (specified in an application context). The situation is
generally more complicated when object behaviour is seen as the
resultant of the interaction of processes at different aggregation
levels. For instance the land use development within an urban
district will be affected the socio-economic developments of the
urban area which it is part of. But the development of the street
block will also be constrained by its inner structure, i.c. the
characteristics of its constituent components. This means that the
behaviour of objects is constrained by processes at a lower and at
a higher aggregation level.
CONCLUSION 5
Object semantics should be understood and specified in multi
aggregation level (multi scale) context.
The examples of this paper referred to 2-dimensional situations.
The conclusions are, however, equally valid for 3-dimensional
approaches.
