ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS", Bangkok, May 23-25, 2001 
5 
Ps and goes across 
into segments. For 
tance between OPs 
e the local distance 
ength to the whole 
, the local skeleton 
in Figure 7. The 
celeton length, k the 
skeleton width. 
on skeleton takes 
atial distribution and 
ling A and B keeps 
ting in the distance 
ition, we can feel in 
eft due to C position 
computation will get 
>serer than that of A 
insistent with visual 
ed for one skeleton 
ind the normal line 
rection between two 
used in next section 
GENERALIZATION 
jping, displacement, 
ing model is able to 
its. From high level 
:tion discusses the 
i detail. 
condition of conflict 
ts : conflict skeleton 
rith weighted width 
lentified as conflict 
ted to one or more 
Iding objects. Figure 
skeleton and conflict 
ifiict skeletons, 
ig displacement 
nd dark dot. 
ject can be assigned 
lacent distance, and 
d other non-distance 
The judgement of conflict building object answers the question of 
who will displace. The further question is how far and what 
direction the conflict building moves. 
If the conflict object has only one conflict skeleton, then the 
adjacent direction serves as moving direction. Otherwise, using 
vector add operation computes the integrated moving direction. 
We suppose each conflict building is attracted by its neighbor 
conflict building and the attraction force is equal. When one 
building is attracted by neighbors from two opposite direction, or 
surrounded by conflict buildings (it means all skeleton related to 
one building are conflicted), it will keep unchanged. In actual 
application, when the added vector length is shorter than a 
threshold, we can think no one direction attraction is strong 
enough over other directions and also regard the object as fixed. 
Figure 9 expresses the movement direction of conflict object, the 
dark arrow symbol representing the displacement direction and 
the dark dot representing the building fixed. 
For offset length of displacement, firstly we suppose the position 
accuracy is not less than half of conflict distance. It means 
conflict building moving face to face and meeting together in one 
position is not against position accuracy. Parallel with the 
displacement direction, draw an extended line from each vertex 
of conflict OP and compute the distance between start vertex 
and intersection point of extended line and GP boundary. Select 
the shortest distance as displacement offset length. This process 
guarantees each building moving within its own GP range, not 
overlapping with other building’s GP. It means the displacement 
will not result in new conflict. This point is very important in 
displacement generalization research (Mackness 1994). 
The purpose of displacement in building cluster generalization is 
to maintain statistic area balance. But generally after 
displacement, it is not yet to get two buildings that exactly share 
a common seamless boundary, still existing gap area. An 
improvement is to execute rotation, but rotation angle and 
rotation scope is complex to decide and yet can not resolve 
problem completely. 
4.4 Progressive Generalization Workflow 
The above sections investigate the decision analysis and 
Generalized result and growth polygon 
4.3 How to Aggregate? 
Considering the square characteristics of building shape, the 
a 99 re 9 ai ion of displaced buildings has to maintain orgothonal 
nature. We use the method of two vertical direction scanning and 
filling on the basis of raster data structure, including 6 steps: 
finding MBR, rasterizing, scanning and filling lines, scanning 
and filling columns, vectorizing, rotating. The whole procedure is 
shown in Figure 10. 
1> finding MBR, 2> rotating and, 
rasterizing 
3>scanning and 
filling lines 
4> scanning, 5> vectorizing 
and filling columns 
6> rotating to 
original direction 
Fig. 10. Six steps of building aggregation 
This method applies two vertical direction of MBR edge to scan 
and fill raster element using the suppose that the MBR edge 
direction can represent the main direction of building cluster. It 
guarantees that the gap area between neighbor buildings is filled 
in the shortest connection. 
Fig. 11. An illustration of progressive generalization