with class hierarchies. In the process of object aggregation 
the information of lower level objects is aggregated to 
higher level objects, but in principle the original detailed 
information is maintained so that it is possible to access 
the detailed information of the lower level objects through 
the aggregated objects. The result of such an aggregation 
process is a less detailed terrain description that may be 
compared to the result of a map generalization. 
The output of this process could be used as the input for 
a following aggregation step. This has been illustrated in 
figure 11, that shows a process starting from the situation 
of figure 4.C. The regions of situation C are assigned to 
more general classes in situation D and the aggregated 
to form the larger regions of situation E. 
CLASS GENERALIZATION STEP 2 OBJECT AGGREGATION STEP 2 
27 = natural grassland 
“SOFT natural vegetation 238 =o & 
38, 52=> forest a? 23 — 2 
2 «c agriculture uw] agriculture 7,52 c==t= 85 
  
   
fig. 11: The second aggregation step for the objects 
of figure 4.C. 
The class generalization and object aggregation steps of 
the approach of figures 4 and 11 have been represented 
in a different way in figure 13. This figure combines per 
database generalization step two steps like those of figure 
4. In the first step of figure 13 the objects of the different 
classes are first assigned to the super classes at the next 
higher level in the hierarchy (compare the class generaliz- 
ation step of fig 4), then in the same step the objects that 
form a region per super class are aggregated to form a 
larger object (compare the object aggregation step of fig 
4). This procedure is repeated in the second step of figure 
13. 
This figure shows that the two steps of the example of 
section 3.1 reduce the number of objects, that is why we 
rather talk of database generalization because the process 
generated objects with a lower spatial and thematic 
resolution then the original objects. Due to the fact that 
the original objects formed a geometric partition of the 
mapped area and due to the fact that generalization 
process made use of the topologic and hierarchical 
structures in which the objects had been modelled, this 
process resulted in a new set of objects that also formed 
a geometric partition of the mapped space. But the result 
was a terrain description of a reduced spatial complexity 
as is shown in the stepwise reduction of the complexity 
of the adjacency graphs of figure 12. 
Each object is represented by a node in these graphs and 
the adjacency between two objects is represented by an 
arc. This figure gives the adjacency graphs related to each 
stage of a process that starts from the situation B of figure 
3 where the original objects have been assigned to their 
552 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
  
fig. 12: Theadjacency graphs related to differentstages 
of the generalization processes of figures 4 and 11. 
super classes. If we follow the steps of fig. 12 then we 
see that: 
- in step a the regions per class have been identified, 
- in step b the objects in each region are aggregated 
to form a composite object that is represented by 
one node, 
- these regions are after step c assigned to more 
general classes, 
- in step d regions at this higher class level have been 
identified, these are composed of the objects 
obtained after step b, 
- then finally after step e each of these regions have 
been aggregated again to form the objects at the 
higher aggregation level which is then represented 
by one node, this is the adjacency graph of situation 
E of figure 11. 
The reduction of spatial complexity is one of the important 
aspects of generalization processes as they are known in 
mapping disciplines. This process has traditionally been 
applied in the form of map generalization to reduce the 
information content of a map so that a mapped area could 
be represented at a smaller map-scale. This process has 
two steps, the conceptual generalization and the graphic 
generalization. The conceptual generalization results in a 
redefinition of the mapped spatial features or objects to 
reduce their number for the terrain description at the smaller 
scale. The graphical generalization isin facta simplification 
ofthe graphical representation of these features or objects, 
including such aspects as geometric simplification, object 
displacement, resymbolization etc. 
10. CONCLUSION 
When we deal with spatial database generalization in a 
GIS environment then this might include the graphical 
representation as well, but that is not necessarily so. The 
main aim will be a simplified terrain description, i.e. alower 
spatial complexity to emphasize spatial patterns and 
relationships that might be difficult to find in a more detailed 
terrain description. That means that this process is very 
much related to the conceptual generalization step 
mentioned before the main aim of this step is to obtain 
a data reduction. We have seen that it can to a large extend 
   
fig. 
be de 
databe 
structt 
will be 
A spati 
aspect 
of sec 
aspect 
of the 
that th 
will c: 
databa 
type o 
should 
origina 
decide 
tion of 
some tL 
interes 
charac 
transfo 
A good 
useful 
acquisi 
a rever 
proces: 
resoluti 
resoluti 
Image i 
where 
general 
objects 
objects 
tion, w 
resoluti 
informa 
informa 
segmer