ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS”, Bangkok, May 23-25, 2001 
concrete operation in building cluster generalization. How to use 
them in a complete generalization process depends on workflow 
control. Considering the fact that conflict in building cluster is 
related to each other, we can not simply aggregate all the conflict 
object which is connective. Aggregation of part of conflict object 
and displacement may resolve the conflict between different part 
groups. Especially when scale changes largely, the predefinition 
of large conflict distance may lead to all building locating within 
one street block are conflict. Obviously it is not proper to 
combine all building into a big one. The whole control workflow of 
building cluster generalization should be a progressive 
procedure to remove conflict step by step. 
If the distribution frequency of skeleton width covers a broad 
range, and the width value is able to be obviously distinguished, 
we can introduce MST method idea(Regnauld 1997) to control 
the generalization procedure. It takes into account the distance 
difference not only in quality between conflict and non-conflict 
but also in quantity. The workflow is described briefly as follows. 
Repeat the following steps until step i> finds no conflict: 
i> Construct triangulation, compute GP and find conflict 
skeleton, conflict building object. 
ii> Sort the conflict skeleton on weighted width from short to 
long. 
iii> Scan conflict skeleton to check the related left and right 
conflict OP. Two OPs can only remain current scanned 
skeleton as conflict. Remove other conflict skeletons. 
iv> Resolve remained conflicts using the above aggregation 
method. 
The above workflow guarantees each conflict removal happens 
exactly between two buildings. Figure 11 illustrates some 
procedures of building cluster generalization. 
If the building distribution is random and the conflict s are few, 
the above workflow can get proper generalized result. The 
questions exist in next two aspects. 
1> The early aggregated building will displace many times in 
following processes and the position accuracy may be 
damaged. 
2> Distribution pattern can not be maintained. 
The workflow improvement depends on further grouping the 
conflict objects which have been identified by adjacent distance. 
The mini distance difference is not able to distinguish building 
group, requesting non-distance standard. The Gestalt nature in 
building size, orientation, shape, distribution structure is an 
important consideration fact. 
Connecting center points within Voronoi diagram polygon gets 
Fig. 12. The network of connective conflict 
building object 
dual geometric construction, Delaunay triangulation. 
Correspondingly, based on building partitioning model, 
connecting representative points of conflict building obtains 
some connective networks, as shown in Figure 12. The further 
works of this research in the future is to discover building 
distribution pattern based on network analysis and combined 
with other methods. 
5. CONCLUSION 
Based on Delaunay triangulation skeleton, this study constructs 
a building partitioning model which is similar to Voronoi diagram. 
The nature of equally separating space makes it a powerful tool 
to analyze polygon distribution cluster. When applied in building 
cluster generalization, it enables to solve conflict detection, 
displacement offset and direction computation. The improved 
distance computation between two buildings takes into account 
the context environment and conforms to visual cognition. The 
model and algorithm presented in the paper has been realized in 
an interactive map generalization system. 
Independent building simplification gets some achievements. 
Building cluster generalization belongs to high level research 
facing challenges. The representation and automatic recognition 
of spatial distribution pattern is the first question to be resolved. 
References 
[1] Ai, T. , Guo, R. and Liu, Y. “ A Binary Tree Representation of 
Bend Hierarchical Structure Based on Gestalt Principles”, 
Proceedings of the 9 th International Symposium on Spatial 
Data Handling, Beijing, pp2a43-56, 2000. 
[2] Ai, T. and R. Z. Guo (2000): A Constrained Delaunay 
Partitioning of Areal Objects to Support Map Generalization, 
Journal of Wuhan Technical University of Surveying and 
Mapping 25(1 ):35-41 ,(in Chinese). 
[3] Bader, M. and R. Weibel (1997): Detecting and Resolving 
Size and Proximity Conflicts in the Generalization of 
Polygonal Maps, Proceedings of the 18 th ICC, Stockholm, 
Sweden, Vol. 3: 1525-1532. 
[4] Christensen, H. J. Albert (1999): Cartographic Line 
Generalization with Waterlines and Medial-Axes, 
Cartography and Geographic Information Science, 
26(1 ):19-32. 
[5] Federico Thomas (1998): Generating Street Center-Lines 
From Vector City Maps , Cartography and Geographic 
Information Systems, 25(4):221-230. 
[6] Guo, R. Z. and T. H. Ai (2000): Simplification and Aggregation 
of Building Polygons in Automatic Map Generalization, 
Journal of Wuhan Technical University of Surveying and 
Mapping, 25(1 ):25-30 (in Chinese). 
[7] Heller, M.(1990): Triangulation Algorithm for Adaptive 
Terrain Modeling, Proceedings of the 4 th International 
Symposium on Spatial Data Handling, Zurich Swiss, 
International Geographical Union, 163-174. 
[8] Hernandez, D. and Clementini, E. (1995): Qualitative 
Distance, Proceedings of COSIT’95,Semmering, 
Austria:45-57. 
[9] Jones, C. B., Bundy, G. L. and J. M. Ware (1995): Map 
Generalization with a Triangulated Data Structure, 
Cartography and GIS, 22(4): 317-331. 
[10] Lee, D.(1999): New Cartographic Generalization Tools, 
CD-Rom Proceedings of 19 ,h ICC, Ottawa , Section 8. 
[11] Mackaness, W. A. and K. Beard (1993): Use of Graph 
Theory to Support Map Generalization, Cartography and 
Geographic Information Systems, 20(4):210-221. 
[12] Mackaness, W. A. 1994. An Algorithm for Conflict