Point-feature label placement with maximum entropy principle

Abstract

A map becomes readable and translatable only after the use of labels. High-quality label placement (i.e., labelling) is a combinatorial optimization problem, where one or more objective functions are required. However, such objective functions are still not well achieved in existing methods with commonly used labelling rules (e.g., “avoidance of overlapping labels” and “placement at priority positions”). In this study, we define a new objective function using the maximum entropy principle and employ a modified genetic algorithm in the optimization process. Comparative experiments have been conducted with 29 existing labelling methods in a dataset of 1000 points. Experimental results show that the average percentage of non-overlapping labels by this new method reaches 94.73% (i.e., being ranked second in all 30 methods) and has no statistically significant difference from the best one (i.e., 94.87%). By further analysis, it is found that this new method not only can place more labels in priority positions than the method being ranked first but also has the potential to place labels for line and area features. Such results indicate that the maximum entropy principle may pave the way for further research in cartography.