Geoscience Reference
In-Depth Information
Stroke-based selection
:
this approach, first proposed by Thomson and
Richardson (1999), had two steps,
build stroke
and
order
stroke
.
1.
Build stroke:
To concatenate continuous and smooth network arcs as a
whole. Several criteria may be used. It was found from our previous
research that, when road class, road name and the every-best-fit (a
geometric criterion named by Jiang and his collaborators (2008)) are
combined together, the built strokes can have a better result. To be
specific, road segments with the same road class are first concatenated
together. If road segments at an intersection are all with the same class,
road names are then used to determine the concatenation. If road seg-
ments at an intersection are all with the same road name or all with dif-
ferent road names, every-best-fit is finally employed to determine the
concatenation.
2.
Order stroke:
After strokes are built, the importance of each stroke can
be calculated according to various properties. And stokes may be
ranked in a descending order. Selective omission becomes as simple as
selecting a number of strokes ranked ahead. And different rankings
result in different selections.
Mesh density-based elimination:
this approach was proposed by Chen and
his collaborators (2009). In this approach, road network is abstracted by
eliminating unimportant road segments rather than selecting important
ones. In a road network, the mesh (i.e. a close region surrounded by
several road segments) with the maximum density is detected. And bounding
road segments of the mesh with the least importance are eliminated. The
importance of each road segment is determined according to the impor-
tance of its own stroke (each road segment can be belonged to a unique
stroke). This processing is repeated until no mesh is left.
3.4 Experimental benchmarks
The aim of this research is for road network generalization, thus the selec-
tion results are compared with maps at different scales. The digital map at
1:20,000 scale is used as source data to which selection omission operator
is implemented. And other smaller scales (1:50,000, 1:10,000 and
1:200,000 scales) produced by the Hong Kong Land Department are used
as benchmarks for evaluation
.
