Geoscience Reference
In-Depth Information
lends itself to be generalized through this special operator: ditches. A de-
tailed description of ditches and the reasons why they require typification
is given in section three. The algorithm will be explained in details in sec-
tion four. Although the algorithm has been developed for the generaliza-
tion at such large scale, our guess is that the procedure is general enough to
be applied also to other (smaller) scales.
2- Related work on typification
Typification and pattern recognition are closely related, as the first step of
typification is to understand the pattern that should be kept.
Pattern recognition techniques have been developed for roads and build-
ings. (Heinzle et al. 2005) find grid-shaped structures in roads analyzing
the nodes of the road graph; (Christophe and Ruas 2002) find alignment in
buildings while (Anders and Sester 2000) developed a parameter free clus-
ter recognition algorithm that can be used as a preprocessing step for typi-
fication. Due to the simple shape of ditches, the pattern recognition is not
as complex as that of roads or of buildings, and thus let us develop a
strategy much simpler that those in these works.
Many papers investigating pattern recognition techniques focus also on
typification. (Regnauld 1996, 2001), uses a minimum spanning tree to
cluster and typify buildings, (Sester and Brenner 2000) developed a typifi-
cation algorithm based on Kohen maps, (Burghardt and Cecconi 2007)
apply mesh simplification to solve the same problem. Most of the work on
typification focuses on buildings, but there are also some examples of road
typification in (Thom 2005), (Luan and Yang 2010)). Neither the works on
the typification of buildings nor those on the typification of roads can be
extended to our case: the algorithms cited in fact operate on geometries
(e.g. polygons) or pursue objectives (e.g. collapsing dual carriageways)
that are quite different from ours. The clustering strategies proposed could
find application in our case, but at the moment the implementation relies
simply on a fixed maximum distance threshold to group ditches.
