Examples of results
The purpose of this section is to provide some result samples of some works on generalisation. These samples concern several different kinds of generalisation (different types of objects, operations, scale changes...). The coordinates of the author(s) are available. If you want the results of your work to be published on this page, contact the chairs.
Automatic knowledge revision in a generalisation system (COGIT)
This work deals with the automatic revision of procedural knowledge for generalisation systems based on the agent paradigm. Our approach consists in analysing the system execution logs and extracting new knowledge from thee logs using machine learning techniques. The figure present an application of our approach on housing estates.
Contact: Patrick Taillandier (Laboratoire COGIT, France)
Road and river network selection (COGIT)
Network selection seeks to choose the features from an initial geo-database that will appear in a generalised geo-database. For road and river network, it is complex process as a good choice should reduce the amount of features but should also preserve the characteristics of the networks (hierarchies, density differences, specific patterns). The automatic processes developed at COGIT are based on a complete enrichment of the initial database by specific structures and patterns of the network (e.g. braided streams for rivers or highway interchanges for roads) to allow a good selection.
Result on a river network
Result on a road network
Contact: Guillaume Touya (COGIT laboratory, IGN France)
Label generalization (ICC)
Results of the automated selection and scaling of map names done by ICC software.
original 1:5000
generalised 1:10000
Contact: Maria Pla and Blanca Baella (Institut Cartogràfic de Catalunya)
Simultaneous graphic generalisation (Lund University, Finnish Geodetic Institute)
This study deals with an optimization (with continuous variables) approach to graphic generalization (essentially the operators simplification, smoothing, exaggeration and displacement). By solving the whole graphic generalization in a single step, both in terms of operators and all types of objects, we provide a sound compromise between the competing goals in the graphic generalization process. The basic idea behind simultaneous graphic generalization is to formulate the requirements of graphic generalization as analytical constraints and then to use these constraints as observations in a least-squares adjustment.
The figure visualizes the effect of simultaneous graphic generalization in a small area. The original objects are shown in gray and the generalized ones in black. (© The National Land Survey of Finland, all rights reserved.)
Contact: Lars Harrie (Lund University, Sweden)
Comparison of processes for road generalization (COGIT)
Empirical comparison of the three processes:
- GALBE: is a first combination of basic tools to create a fully automated process for independent road generalisation.
- CartoLearn: explores the interest of using machine learning techniques to acquire knowledge necessary to guide generalisation.
- AGENT: uses multi-agent principles and a constraint based approach to represent user needs.
Contact: Sébastien Mustière and Cécile Duchêne (Laboratoire COGIT, France)
Displacement cost (University of Glamorgan)
Example of the effect of including displacement cost.
Original data.
Result obtained without consideration to displacement cost (original object locations shown in background).
Result obtained when displacement cost is taken into account; unnecessary displacement has been reduced.
Contact: J. Mark Ware (School of Computing, University of Glamorgan)
Hydrography and Transportation Generalization (Canada Centre for Remote Sensing)
Original transportation and hydrographic data set for an area in South Eastern Quebec, Canada.
Originally at 20000 (not to scale)
Result after highly automatic generalization suitable for 1:250000 representation (not to scale)
Comparison with scanned manually produced map 1:250000 (not to scale)
Contact: Dianne Richardson (CCRS, Canada Centre for Remote Sensing)
Road displacement with elastic beams (University of Zurich)
Road displacement with elastic beams enhanced for road networks and intersections.
Situation before displacement...
...and after displacement
Intra-line symbol conflicts (top row) can be solved by widening bends, here computed by a beam-based algorithm (bottom row)
Contact: Mattias Bader (Departement of Geography, University of Zurich)
Typification of settlement areas based on Self-Organizing Maps (SOM)(University of Hannover)
Typification
Reducing information content while preserving overall spatial structure (no arbitrary reduction of number of objects!)
Algorithm
given:
- input space E of dimension m with training vectors (stimuli) x
- map space A of dimension d with connected neurons (dimension d is typically 1 or 2)
Every neuron in the map space is described by the tuple U=(w,p) i.e. a weight w in E and a position p in A
Weights of neurons correspond to positions in feature space, that are iteratively adjusted to the training vectors
All neurons are connected Ð> change of one neuron results in changes of its neighbors
Application to typification
Input space: original objects
Map space: subset of original objects, random selection according to reduction rate
Map structure: Delaunay triangulation of subset
Advantage
No determination of spatial distribution necessary as this is implicit in the algorithm
Original
Reduction to 50%
... to 30%
... to 20%
Contact: Monika Sester (Institute for Cartography and Geoinformatics, University of Hannover)
Cartographic Displacement by Minimization of Spatial and Geometric Conflicts (Federal Agency for Cartography and Geodesy)
Based on a physical model this approach describes on the one hand the persistence behavior of the map objects to their original location and shape by simulated suspension of elastic springs. Changing the geometry of map objects will induce an internal potential.
On the other hand the spatial conflicts or external potential is described by intersecting signatures and buffers around these object-shapes. The external potential is determined by the overlapping area and importance of the signatures. The optimal solution is found by minimization of the displacement potential which consists of the sum the internal and the external potential.
Original data with displacement buffer
After optimization
Displaced data
Comparison with original data
Detail of a Topographic Map 1:25000 (TK25) before (left figure) and after (right figure) displacement
Contact: Joachim Bobrich (Federal Agency for Cartogrpahy and Geodesy)
Page authors: Martin Galanda and julien Gaffuri
If you have pictures to send to complete this section, contact the webmaster.
Chairs: William.Mackaness(at)ed.ac.uk and Sebastien.Mustiere(at)ign.fr