a case study of combined text and icon placement

MethodThe test is performed by repeating text placement where the data shown in Figure 2 was used. In the test we were onlyinterested in comparing methods for choosing candidate positions; therefore, we only run the first three steps in the process(i.e., no deletion was performed). In all the executions the following weights were used (cf. Eq. 1; see Zhang and Harrie foranalytic definitions of preference, disturbance and overlap):* Preference 0.1* Disturbance 0.3* Overlap 1.0Furthermore, all the parameters to control the simulated annealing were the same in all executions (e.g. if the objectivefunction was less than 1.8 the iteration was ended).Figure 2: An example of text label and icon placement using 8 randomly chosen candidate positions. The value of theobjective function was equal to 1.72 (using the weights given above).We tested the following sets of candidate positions:* Number of candidate positions: 4 – Random selection* Number of candidate positions: 8 – Random selection* Number of candidate positions: 4 – Stratified selection* Number of candidate positions: 8 – Stratified selectionFor each set of candidate position we made four executions. Since simulated annealing contains a random process the resultis not the same a set of candidates.

ResultThe result of the case study is presented in Figure 3.Number of candidate positions: 4 Number of candidate positions: 8Random selectionRandom selectionNumber of candidate positions: 4 Number of candidate positions: 8Stratified selectionStratified selectionFigure 3: Result of the case study. On the Y-axis the objective function (Equation 1 without the label removal part) is shownand on the X-axis the execution time (ms). The test was performed on a PC with Pentium IV, 3.0 GHz processorDISCUSSIONA weak relationship is indicated if we look at figure 3 column-wise. An interpretation is that smaller number of candidatepositions would more likely result in faster map labeling (since the optimization process has fewer positions to examine).However it is harder to make any real conclusions based in this result; e.g. the stratified selections with eight candidatepositions has a better execution time than the same selection with only four candidate positions three of four times.Figure 3 also indicates that the random selection gives lower value of the objective function when four candidate positionsare used, but that a stratified gives better result when 8 candidates were used. This is somewhat surprising for us since wethought that a stratified selection ought to improve the chance to find a solution with less overlaps between the text labelsand icons (disregarding the number of candidate positions). However, given our test set up, a random selection seems to bepreferable for 4 candidate positions.

The case study presented here is the first test of evaluate methods to choose candidate positions. To make better conclusionsthe test material must be more extensive.CONCLUSIONSThe aim of the study is to evaluate methods to selecting candidate positions for the text labels and the icons before thecombinatorial optimization. A case study was performed where either 4 or 8 candidate positions were chosen; thecandidates were chosen either random or by a stratified approach. The case study indicates that, for our test set up, a largenumber of randomly selected candidate positions give the worst result. It is the most time-consuming approach even thoughthe value of the objective function is acceptable. Another indication from the test set up is that a random selection ofcandidate positions gives (a little bit) better map quality (value of the objective function) than a stratified selection.ACKNOWLEDGEMENTSThis project was financed by the International Office and the GIS Centre at Lund University, and the GiMoDig project, IST-2000-30090 (which is funded by the European Union via the Information Society Technologies (IST) Programme). Thestudy is also partly supported by the National Science Foundation of China, under grant No. 40101024. We would like tothank our colleagues in the GiMoDig project. Data for the case study were kindly provided by the National Land Survey ofFinland.REFERENCESChristensen, J., Marks, J., and Shieber, S, 1995. An empirical study of algorithms for Point-label placement, ACMTransactions on Graphics, Vol. 14, No. 3, pp 203-232Dörschlag, D., Petzold, I., and Plűmer, L., 2003. Placing objects automatically in areas of maps. Proceedings of the DurbanICC 2003, South Africa.Harrie, L. and Johansson, M., 2003, Real-time data generalization and integration using Java. Geoforum Perspektiv, Februar2003, pp.29-34.Harrie, L, Stigmar, H., Koivula, T., and Lehto, L., 2004. An Algorithm for Icon Placement on a Real-Time Map. In: Fisher,P., 2004. Developments in Spatial Data Handling, Springer.Imhof, E., 1975. Positioning Names on Maps. The American Cartographer, Vol. 2, No. 2, pp. 128-144.Petzold, I., Gröger, G., and Plűmer, L., 2003, Fast Screen Map Labeling – Data Structures and Algorithms. In Proceedingsof the 21st International Cartographic Conference (ICC), 10 - 16 August 2003, Durban, South Africa, pp.288-299.Russell, S., and P. Norvig, 1995. Artificial Intelligence – A Modern Approach, Prentice-Hall.Strijk, T., and van Kreveld, M., 2002, Practical extensions of point labeling in the slider model. GeoInformatica, 6:2,pp.181-197, 2002.van Dijk, S., 2001. Genetic Algorithms for Map Labeling. PhD thesisT, Utrecht University.van Kreveld, M., Strijk, T., and Wolff, A., 1999. Point Labeling with Sliding Labels, Computational Geometry, Vol. 13, pp.21-47.Vivid Solutions, 2005. Java Topology Suite, http://www.vividsolutions.com/jts/jtshome.htm (accessed 2005-04-21).Wolff, A., Knipping, L., and van Kreveld, M., 1999. A simple and efficient algorithm for high-quality line labeling. In:Proceedings 15th European Workshop on Computational Geometry (CG'99), pages 93-96, Sophia-Antipolis, 15-17 March 1999.Wolff, A., 2005. The Map-Labeling Bibliography, http://i11www.ira.uka.de/~awolff/map-labeling/bibliography/ (accessed2005-04-21).Zhang, Q and Harrie, L., 2005a. Real-Time Map Labelling for Mobile Applications. Computers, Environment and UrbanSystems, Accepted.Zhang, Q and Harrie, L., 2005b. A real-time Method OF PlacING text and Icon labels simultaneously. Submitted.Zoraster, S., 1997. Practical Results Using Simulated Annealing for Point Feature Label Placement. Cartography andGeographic Information Systems, 24:4, pp.228-238.

A BIOGRAPHY OF THE PRESENTING AUTHORLars has a Master of Science in Geodesy from Royal Institute of Technology, Stockholm, Sweden (1993). Since 1994 hehas been active in Cartography and GIS at Lund University, Sweden. He completed his PhD 2001, with a thesis aboutoptimisation techniques in cartographic generalisation. Is currently a post doctoral fellow at Lund University.