International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
called scale-less or scale-free with a single maintenance procedure
(Muller, 1995) that is appropriate for multi-purpose applications. In
this regard, derivative mapping is considered the most cost effective
and efficient method to derive multiple scale maps and GEODATA,
from a master detailed database to satisfy the map content
requirements of a specific application (Kazemi, 2003).
A major advantage of database generalization is reduction in the cost
and workload of the manual process once the database is highly
structured and attributed. It selects the set of features or attributes and
chooses an approximate level of generalization, then employs
generalization operators/algorithms, and finally post-processes the
dataset. This limits the degree of human intervention. However,
cartographic — generalization renders — the features for
display/visualization as a means of communication that is a
subjective process since cartographers should satisfy the basic
requirements for graphic clarity and legibility, as well as analyzing
the relevance of map features, including their geometric and semantic
attributes by applying generalization operations (Meng, 1997).
Furthermore, reducing the number of features in database
generalization is a key task that can be accomplished by six major
operations based on the geometric, semantic relationships, and
database constraints that are well documented in the literature (e.g.
MeMaster and Shea, 1989 and 1992). These include simplification
(line generalization), aggregation (combination geometrically and
thematically), symbolization (for line, polyline and point), feature
selection (elimination and delete), exaggeration (enlargement) and
displacement or moving objects (Oosterom, 1995).
In cartographic generalization a cartographer chooses features from a
larger scale map to be shown on a smaller scale map through
modifications to filter out detailed information, while maintaining a
constant density of information by considering purposes of the map
(Davis and Laender, 1999). A drawback of this approach is that this
generalization is based on a cartographer's skills, including his/her
visual/aesthetic sense (e.g. clarity, readability, ease of interpretation)
and the lack of extensibility for multiple representations in GIS. The
way forward is incorporation of a data modelling process, as it
provides a detailed description of the database structure, or the so-
called schema. A main advantage of this approach is a reduction in
spatial and semantic resolutions, which permits both a spatial
analysis and map production. For example, Jiang and Claramunt
(2002) proposed a generalization model of an urban street network
that aims to retain the central structure of a street network by relying
on a structural representation of this data employing graph modelling
principles (e.g. Gross and Yellen, 1999). The proposed method
provides a flexible interactive solution to a street network because it
incorporates the concept of a hierarchy-based generalization in terms
of connectivity to an average path, length and measures. Peter and
Wiebel (1999) identified several measures, such as size, distance and
proximity, shape, topological, density, distribution, pattern and
alignment, as a set of generic constraints that need to be applied to
database generalization. Therefore it is suggested that selection of
appropriate algorithms and prioritizing of constraints needs to be
studied. A series of selection rules emerged for road networks, such
as if the average segment length of a street is less than a given
threshold, then keep it in a database, otherwise delete it.
Shortcomings of the graph theory approach are geometric aspects of
coalescence as well as imperceptibility, and semantic perspectives
(e.g. avoiding large detours that are not clearly explained). Kreveld
and Peschier (1998) used this method for road network
generalization.
In this regard, multi-resolution spatial databases provide the ability to
represent objects in multiple representations tailored towards the
requirements of different users, especially for web applications. It
should preserve spatial relations throughout scale changes (Tryfona
and Egenhofer, 1996). Generally there is a direct linear relationship
between scale changes and the amount of generalization (Kerveld,
2001). Continuous map scale change is already operational on
modern computers and so technological developments will soon
provide this capability for web cartography (Karaak and Brown,
2001; cited by Kerveld, 2001). With reference to linear features, note
that a consistent representation of networks such as roads needs to be
ID
ID
considered through two major criteria, including: (a) when small
changes take place from the one level to the next, and (b) when long
changes accrue (Tryfona and Egenhofer, 1996). An example of such
change is presented by Kazemi (2003). Ibid (2003) commented on
the potential for developing a conceptual model for an object-
oriented continuous master database, multi-scale and multi-purpose
database that enables derivative mapping.
3.0 GENERALIZATION OPERATIONS
There are no standard definitions for generalization operations, and
each researcher has defined them based on his/her perspectives or
application area (Cecconi et al, 2000). However, McMaster and
Shea (1992) defined twelve operations, with special emphasis on
digital cartography, with the first ten operators based on graphical
representations, with the last two operators are attributes of the
spatial objects. The applications of some of these operations define
generic rules, whereas some are just used by cartographers in a
subjective manner (Davis and Laender, 1999) due to different feature
type geometry. Thus, a definition of each of the operators could have
different meanings in terms of the feature type, e.g. area elimination
of vegetation features and elimination of hydrographic features. Lee
(1993) examined operational consequences and developed criteria
through formalizing workflow using the MG Integraph software
product for generalization of areal, linear and point features. Results
are presented at 1:100,000 scale, by which the amount of information
kept in the final map was comparable to the real work. Again, there is
no holistic or even ideal sequence for the utilization of these
operations. However, Monmonier and McMaster (1991) claimed
there are sequential effects of the operations in cartographic line
generalization, but have not received much support from others as
each of the operations may serve a specific generalization problem.
Typically the intention is to break down the generalization process
into sub-processes, and later combining several operators to build a
more robust generalization workflow. Also Cecconi ef al.. (2000)
evaluated and integrated generalization operations to improve
automated generalization for on-demand web mapping trom multi-
scale databases. This is an excellent example of recent work on
combining existing generalization algorithms for an operational
environment for on-the-fly map generalization. To date commercial
GIS tools have incorporated many of these operations, but some of
these operations (e.g. displacement, exaggeration) are still in an
experimental form since they are strongly based on a cartographer's
intuition. For example, ESRI's recent object-oriented ArcGIS
software (version 9.0) provides a spatial framework to support
eneralization needs by introducing geoprocessing concepts and map
:eneralization tools that have been enhanced and implemented in a
c
geoprocessing framework (Lee, 2003).
as
Typically current GIS software applications offer both line
generalization and area generalization algorithms. Since the focus of
this rescarch is on road network generalization, this paper only
highlights some of relevant literature on linear features (Skopeliti and
Tsoulos, 2001). Linear feature generalization plays an important role
in GIS (Barrault, 1995; Forghani, 2000; Skopeliti and Tsoulos, 2001).
Several algorithms have been developed to simplify lines. McMaster
(1989) classified the processing of linear features into five major
algorithmic categories: (a) independent point algorithms of map
generalization where a mathematical relationship between
neighbouring pairs of points is not established; (b) local processing
routines that apply the characteristics of immediate neighbouring
points to determine selection; (c) extended local processing routines
that apply distance, angle, or number of points to search beyond
neighbouring points; (d) extended local processing routines that use
morphologic complexity of the line to search beyond neighbouring
points; and (e) global routines that take into account the entire line or
specified segment. However, none of these methods leads to an
automated generalization mechanism.
One of the revolutions in generalization was the development of an
algorithm by Douglas and Peucker (1973) and Duda and Hart (1973)
(iterative endpoint fit). This algorithm is regarded by many as the
best of the line generalization algorithms incorporated into GIS tools
(e.g. Visvalingham and Whyatt, 1993). It should be noted that the
underlying concept of the Douglas and Peucker algorithm comes
Interna
from Al
curvatur
research
algorithi
Whyatt,
approxir
criticize
introduc
is applic
is still,
generali
A majori
are bour
methodo
segment:
for the c
stated tl]
compute
acceptab
ongoing
properly.
cartograp
in an obj
requirem
Whyatt,
existing
generaliz
applicatic
features.
generaliz
order to g
250K Se
1:1000,0€
4.0 GENI
An excell
on measul
is provide
presented
data to ac
into asses
transform:
measures
objects. T
initial ass
identificat
the gener:
evaluation
review of
constraints
generalizat
In conne
categorize
topologica
(object le
objects/reg
relevant to
size (gray
(graphical)
intersection
(Structural)
for linear fi
1999: Yao
Classes) ir
distribution
(3001) di
neighbourh
aggregabili
alignment/T
detailed de:
Peter and V
and Jiang ai