Lan, Li, Peng and Gong – AUTOMATED LABELLING OF SCHEMATIC MAPS BY OPTIMIZATION WITH KNOWLEDGE ACQUIRED FROM EXISTING MAPS 2020

Description

This study addresses automated labeling of schematic metro maps by learning placement rules from existing maps and integrating them into an optimization algorithm. Because many cartographic rules are hard to formalize, the authors extract potential label positions and their preferences from 20 representative octilinear metro schematics, modeling 255 possible station–edge situations and ranking candidate positions by frequency. They embed these learned rules in a local optimization that (1) places initial labels at highest-preference positions, (2) identifies labels with overlaps using a weighted cost (w1,w2,w3; empirically w1=3,w2=1,w3=1), and (3) iteratively moves conflicting labels to lower-preference candidates to minimize overlaps among labels, points, and edges. Implemented in C#, the method was evaluated on Hong Kong and Tianjin networks (~93 and ~150 stations) via a user study (n=27) measuring ease of finding labels, congestion, and satisfaction. Results show comparable or better congestion and ease scores versus official maps, though handcrafted maps scored higher on aesthetic satisfaction. Limitations include the local optimizer and limited aesthetic modeling; future work may explore global or AI-based learning.