ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS", Bangkok, May 23-25, 2001 
ISPRS, Vol.3‘ 
3.4.1 Distribution density 
For generalization application, the building distribution density 
supports operation decision from macro perspective. In 
partitioning model, the description of building distribution being 
dense or sparse depends on the rate of OP area to GP area. The 
denser, the smaller room one building gets during competing 
outward, and the more similar of GP size to OP size. So this 
area rate ranging from 0 to 1.0 is able to represent building 
distribution density. Figure 5 is an example of this kind of density 
representation. The region is shaded with gray scale proportional 
to the rate of OP area to GP area. 
Fig. 5. An illustration of representation of building 
distribution density, the region shaded with grey 
scale proportional to the rate of OP area to GP area. 
3.4.2 Adjacent Distance 
In building cluster, the adjacent degree between two buildings 
can not simply be described as minimum distance, such as in 
Figure 6. What it means for A to be near B depends not only on 
their absolute positions(and the metric distance between them), 
but also on their relative sizes and shapes, the position of other 
objects, the frame of reference (Hernandez and Clementini, 
1995). The context environment plays an important role. 
H5~| 
> 
> 
i d 
A 
Fig. 6. The minimum distance is same, but C is 
closer to A than B in visual cognition. 
Based on partitioning model, applying differential idea we give 
the following method to compute distance between two buildings. 
k 11P; R +i " 
w- 
s 
IWiVfcl 
Type I Type II Type III 
Fig.7, W,W 2 , skeleton width representation for 3 
types of triangle 
Fig 8. Neighbor C moving and rotating leads to the 
distance change between building A and B based 
on the weighted skeleton width computation.. 
A GP boundary edge relates two adjacent OPs and goes across 
a set of triangles which divide the skeleton into segments. For 
each short segment, compute this local distance between OPs 
according to triangle type and then integrate the local distance 
weighted with the rate of local segment length to the whole 
skeleton length. For three types of triangle, the local skeleton 
width representation, WiW 2 is expressed in Figure 7. The 
computation function is where / the whole skeleton length, k the 
number of involved triangle, w is also called skeleton width. 
This weighted distance computation based on skeleton takes 
into account the building shape structure, spatial distribution and 
other building’s influence. In Figure 8, building A and B keeps 
unchange and C moves and rotates, resulting in the distance 
between A and B decreased. In visual cognition, we can feel in 
the right A and B are closer than that in the left due to C position 
change. For Figure 6, this weighted distance computation will get 
the adjacent relation between A and C is closerer than that of A 
and B. So this distance computation is consistent with visual 
cognition in some degree. 
3.4.3 Adjacent Direction 
An approximated direct line can be computed for one skeleton 
using Least Square Adjustment method, and the normal line 
direction can be regarded as the adjacent direction between two 
adjacent buildings. Adjacent direction will be used in next section 
for building displacement. 
4. APPLICATION IN BUILDING CLUSTER GENERALIZATION 
Building cluster generalization involves grouping, displacement, 
aggregation and simplification. The partitioning model is able to 
support the generalization in several aspects. From high level 
decision to low level operation, this section discusses the 
process on the basis of many experiments in detail. 
4.1 Where Is There Conflict? 
The weighted skeleton width acts as the condition of conflict 
recognition. We present two conflict concepts : conflict skeleton 
and conflict building. Those skeletons with weighted width 
shorter than predefined tolerance are identified as conflict 
skeletons, and those building objects related to one or more 
conflict skeletons are defined as conflict building objects. Figure 
9 gives an example of judgement of conflict skeleton and conflict 
building. 
9. Experiment illustrations of conflict skeletons, 
visualized as wide line, and building displacement 
direction, visualized as arrow line and dark dot. 
According to GP connectivity, the conflict object can be assigned 
into classes. This class is depend on adjacent distance, and 
further grouping needs Gestalt analysis and other non-distance 
assessment. 
The judgemei 
who will disp 
direction the ( 
If the conflicl 
adjacent dire< 
vector add op 
We suppose 
conflict buildi 
building is att 
surrounded b; 
one building 
application, v 
threshold, we 
enough over < 
Figure 9 expr< 
dark arrow sy 
the dark dot r< 
For offset len< 
accuracy is t 
conflict buildir 
position is n 
displacement 
of conflict OF 
and intersects 
the shortest d 
guarantees e; 
overlapping w 
will not resull 
displacement 
The purpose < 
to maintain 
displacement, 
a common s 
improvement 
rotation scope 
problem comp 
4.3 How to A< 
Considering t 
aggregation c 
nature. We us 
filling on the 
finding MBR, 
and filling coli 
shown in Figu 
1> finding 
4> scannin 
and fili 
This metf 
and fill re 
direction 
guarantee 
in the she 
4.2 How to Displace?