The International Archives of the Photogramme try, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
293 
matching of end point need infer form the node matching. The 
strategy of matching is shown in figure 4. 
For 1:N route matching, the buffers around every edge need to 
be created firstly. Secondly according to the matching for end 
points of an edge, the edges corresponding are joined within 
given buffer, and then the candidate route matching is found. 
Lastly, the 1:1 route match will be identified by comparing 
measures including length, Huausdorff distance and semantic 
attributes of the routes. Accomplished 1:1 route matching, the 
M:N route matching is processed in succession. The unmatched 
edges on the data at small scale are to be joined into a route at 
every node based on the principle of perceptual organization, 
which describes two edges with the same direction at the 
intersection. So M:N route matching will transform into the 
matching of 1 :N. The method of 1 :N route matching mentioned 
above can be carried out again. 
Figure 4. Strategy of matches of objects 
Accomplished route matching, the unmatched edges on the data 
at large scale are filtered. The matching between node and edge 
can be easily inferred according to its definition mentioned 
above. 
4. SELECTIVE OMISSION BASED ON MESH AND 
MESH DENSITY 
For the selective omission of roads, various techniques already 
exist. For example, the graph theory principles (Mackaness & 
Beard, 1993), space syntax (Mackaness, 1995; Jiang & 
Claramunt, 2004) and self-organizing maps (Jiang & Harrie, 
2004) are employed to support generalization of road network. 
Unfortunately, all these methods failed to consider the 
distribution density of roads which is an important constraint. 
Hu (2007) reveals that selective omission of roads should be 
retained the density difference besides topological, geometric 
and semantic properties of the road network. Two distinctive 
types of density variations are considered, with one across 
different regions within the same map (e.g. urban and rural 
areas) and the other across different map scales of the same 
region. The density conventionally defined as the ratio of total 
length of roads in a given region to the area of the region. 
However, this measure is not sufficient for the purpose of map 
generalization because the local variations of road density over 
the space are not indicated. So a novel approach of selective 
omission for roads is used in this study, which is based on mesh 
and mesh density in a network. 
4.1 Concepts of mesh and mesh density 
A mesh is defined as a naturally closed region that does not 
contain any other region in a road network. Naturally, one 
could consider using area of an arbitrary mesh in the new 
measure for road density. It is called mesh density. In fact, 
mesh density is a special case of the road density expressed 
conventionally and can be described as follows: 
D = P/A 
(1) 
Where P is the perimeter of a mesh, A is the area of the mesh, 
and D represents the mesh density. Form this equation we can 
find that the road stubbles within the mesh boundary are all 
ignored. The road segments in a network can be classified into 
two types, i.e. the segments that constitute mesh boundaries and 
the stubbles lying within the boundaries. Road stubbles can be 
relatively easily handled according to their geometric and 
semantic properties. The elimination of any road stubble will 
not influence the connectivity of network. Moreover, if a mesh 
needs to be simplified, the road stubbles within the mesh will be 
likely omitted prior to boundary segments. For these reasons, 
this simplification is adopted. 
The meshes can be classified based on the types of roads on its 
boundary. A mesh may be bordered by roads of different 
classes such as main streets and secondary streets. Then meshes 
are classified based on the bounding road segments. In this 
method, the class of a mesh is assigned with bounding segment 
of the lowest ranking. For road generalization based on the 
mesh density, a mesh can be regarded as the basic unit of 
selection. With a given set of roads, the mesh density can be 
mapped out. The mesh density is then used as a constraint to 
determine which meshes should be treated. Usually, a threshold 
is given beforehand or computed from the given set of data. 
Moreover different thresholds may be applied to different 
classes of meshes in order to preserve the density variations 
across different regions of a road network. If the density of a 
mesh goes beyond the threshold, then there is a need to 
eliminate one or more road segments on its boundary. 
4.2 Determination of thresholds for mesh density 
There are two different ways to determine a density threshold 
for each class of meshes. They are respectively based on 
theoretical analysis, map specifications and empirical study. 
For maps at a certain scale, there must be a minimum size for a 
mesh unit below which the mesh cannot be perceived anymore. 
Correspondingly, there must be a threshold of mesh density 
beyond which one or more segments on the boundary of the 
mesh must be eliminated so that two or more adjoining meshes 
could be merged to form a larger mesh. Such a threshold is 
regarded as the pennissible largest density (PLD), which 
implies the longest possible length of roads in the smallest 
visible area. The minimum mesh size could be set to the SVO 
(Smallest Visible Object) in “natural principle” proposed by Li 
and Openshaw (1992, 1993). SVO could be a small circle, a 
raster cell or any other geometric entity. So the PLD i.e. density 
threshold can be expressed in terms of the ground size and the 
principles of the SVO as follows: 
D,h = 4/(5,-i„ (l-SJS,)) (2)