城市点与路网集成的小比例尺制图综合

2. Iterate through Lj and delineate clusters using triangulation of the points and merge the adjacent triangles in which two edges are shorter than threshold Tj. At each level of clustering a set of polygons is produced.

3. Add the largest clustering level, which covers all the points.

4. From every polygon subtract all polygons of smaller clustering tolerance.

5. Buffer the resulting polygons by the distance d=0.5Tj

6. Define the percentage Rj of the points to be removed for every clustering level.

7. For every point define the smallest clustering level to which it belongs.

8. For every point calculate Wi according to clustering level; set Ni equal to zero.

9. Iterate through clustering levels.

a. Iterate through cluster polygons at current level and omit the points using the following strategy:

i. Select the points belonging to current polygon and current clustering level.

ii. Derive Voronoy diagram for these points and clip it by current polygon.

iii. Sort the points by Ni (first) and Wi (second) fields in increasing order.

iv. Iterate until the percentage is achieved.

Select the first point in the list and remove it.

Increment Ni of adjacent Voronoy neighbours by one.

Rebuild and clip the Voronoy diagram.

Recalculate Wi for adjacent Voronoy neighbours.

Sort the points by Ni (first) and Wi (second) in increasing order.

Our algorithm was programmed using Model Builder for ArcGIS Desktop 10. Experiments showed that it is flexible, allowing to find an optimal balance in preserving the point pattern, retaining points in sparsely populated areas and rarefying extremely dense regions for labeling and symbol placement (Figure 1). The balance can be controlled by selection of different number of clustering levels Lj, different thresholds Tj and point removal percentages Rj for each level. Resulting values of neighbour index Ni (which is incremented each time the neighbour of the point is deleted) show generalization cores. The points with high value of order are located in intensively generalized regions.

Figure 1. The result of city points generalization using polygonal clusters. Selected

points are shown in blue and clusters in yellow-red gradations.