Geoscience Reference
In-Depth Information
preserves the pattern, requires skills as abstraction and creativity that
belong more to the human than to the machine.
If developing generalization algorithms could be regarded as teaching the
computer how to draw maps, this is particularly true about typification. In
their work (McMaster and Shea 1989) state that to generalize we need to
know why, when and how to generalize; we can write that to develop a
good typification algorithm we need to teach the computer what to draw,
how to draw and where to draw it.
Despite the complexity of such a task, developing a typification algorithm
is not impossible. The following section of this paper contains a brief re-
view of some works on typification: the algorithms described there are a
good example on how constraining the type of patterns to search for and
focusing on just some cartographic phenomena it is possible to develop
smart and effective typification algorithms.
Typification is very useful -and much used- when the conflicts for space
are very high, i.e. when generalizing at small scales, thanks to its ability to
replace a group of objects with a smaller group of objects. At large scales
there is usually enough space to preserve patterns simply reducing the
symbol size or deleting some of the objects in the pattern; in most cases a
selection algorithm, maybe driven by a pattern recognition algorithm,
could be enough to generalize a pattern: the number of objects left will be
sufficient to convey the information about the original distribution. A typi-
cal example of this is the typification of buildings in settlements: at large
scale buildings inside a city block can be generalized simply deleting the
smallest of them; if the spatial structure has to be enforced it is enough to
delete the buildings outside the main direction of the pattern (e.g. a build-
ing not facing a street). At smaller scales, when the space for the represen-
tation of the block is much less, buildings can not be simply selected: the
original buildings should be replaced by a new representation counting
fewer objects through a typification operation.
The algorithm described in this paper has been developed in the frame of
the CARGEN project, a research started in 2006 as a collaboration
between the Department of Information Engineering of the University of
Padua, the local government Regione Veneto and the Italian NMA, the
IGM -Istituto Geografico Militare, with the aim to investigate the automatic
generalization of the 1:25000 IGM geodatabase from the regional geodata-
base in scale 1:5000. Despite the scale at which the CARGEN project
works does not usually require typification, we found a feature class that
