Label placement
Rules, techniques
Labels on a map
• Text, name of map features
• No fixed geographical position
• Labels of point features (0-dim), line features
(1-dim), areal features (2-dim)
Labels of point features
• Cities on a small-scale map
• Heights of peaks
• Measured values at measurement points
Labels of
point features
Touristic sights,
campgrounds
Labels of line features
• Rivers
• Heights on contour lines
• Streets
Labels of areal features
• Countries, oceans, forests, lakes
Three kinds of labels
Utrecht
Nieuwegein
Houten
Culemborg
Straight,
horizontal name
Curved name
Curved name
with spacing
between
characters
Label placement
• Where should the text be placed? Which shape?
• 20% - 50% of the map production time;
can be automated
• Which rules do cartographers use?
- No overlap  readability
- Clear for which feature  association
- Aesthetical guidelines
Rules according
to Imhof (1962)
Problematic cases
Problem association
Thunder Bay and
point
Problem visibility
highway 20
Problem
readability name
Another problematic case
University Ave. is interrupted 4 times
Specific rules point labels
• Label must be close to point, preferably to the
right and above
• No overlap with other labels, except perhaps
in the spacing of the areal label
• Label may intersect line feature; then line
must be interrupted
• Points close to large body of water must have
their label in the water. In other cases there
may be no line between a point and its label
• If it cannot be avoided, a name may be split
over two lines of text
Specific rules line labels
• Label must follow shape of river
• Label should not bend upwards and downwards
consecutively
• At long line features the label must be
repeated
• No extra spacing between characters, spacing
between words of the label is allowed
• For vertical line features: upward reading
direction left on the map and downward right
on the map
• Contour lines: labels must interrupt contour
line; top label points to higher regions
Specific rules areal labels
• Horizontal labeling is good unless this conflicts
with the dominant shape of the region
• Monotonous curving; no inflection points
• Label must be spread over whole region
• Non-horizontal labels must be curved
• Adjacent regions preferably have same shape
of label (both horizontal, e.g.)
• Sometimes an areal label may be outside its
region, but not in another labeled region
Strategies for automatic label
placement
• Compute for each feature various candidate
positions according to the rules
• Choose for each feature one candidate position,
such that the chosen positions don’t have
overlap
• Sometimes a feature cannot be labeled, and
sometimes the label of a feature must be
repeated
Candidate positions points
• Point labels
8-position model
4-position model
slider model
More general model 
more labels can be placed,
but more complex/
expensive to compute
Candidate positions lines
• Compute strip with character height above the
river which follows the shape approximately, but
doesn’t have too high curvature (e.g. with
opening, closure)
• Determine suitable segments in the strip as
candidates
Candidate positions regions
• Determine skeleton (medial axis) of region
(polygon)
• Determine longest path on skeleton as dominant
shape
• Choose middle with suitable character spacing
• Vary a little for more candidates
Candidate positions regions,
horizontal
• Choose spacing and determine length : height
of label
• Compute largest scaled copy of the label
(rectangle) that just fits inside the polygon
• Put the label in the correct font size in the
middle of the scaled copy
B e l g i ë
B e l g i ë
B e l g i ë
Removal part of the candidates
• Compute those candidate positions that
intersect map features significantly, or would
give unclear association, and remove them
Nieuwegein
Nieuwegein
Nieuwegein
Reasons for removal
Utrecht
Utrecht
Zeist
Utrecht
Utrecht
Readability
text
Utrecht
Zeist
Visibility
features
Utrecht
Association
Choosing from the candidates
• Translate to a graph problem (graph G):
- each candidate position of each feature is
a node in G
- two nodes have an edge in G if their
candidate positions intersect
- two nodes have an edge in G if their candidate
positions are of the same feature
Example
Utrecht Utrecht
Bunnik Bunnik
Utrecht Utrecht
Bunnik Bunnik
Computing intersecting
candidates
• Each candidate is (approximately) a simple
geometric shape
• Determine all pairs of intersecting candidates
using a plane sweep algorithm
• O((n+k) log n) time, with k the number of
intersecting pairs
• Typically: k = O(n)
Graph problem: maximum
independent set
• Also called maximum stable set
• NP-hard problem: solving optimally is very time
consuming
• Cannot be approximated with performance
guarantees
• Many heuristics
Heuristics for maximum
independent set
• Select each candidate that is adjacent to a
clique, and remove the candidate and clique
from the graph
• Repeat as long as possible (optimal selections)
• Select candidate that intersects few other
candidates
• Select candidate if a feature has only one
(or a few) candidates left
• Eliminate candidates that intersect many
candidates
Refinements
• Features and labels have different importances
 translate problem to maximum weight
independent set
• First place areal labels, then determine
candidate point labels and select, then
determine line labels and place them
(order of decreasing freedom of placement)
Labels in GIS
• Can label positions be pre-computed and
stored, for better efficiency?
No, because
– Which themes are shown is not known
beforehand;
– Which scale and size of the map is needed is
not known beforehand
Labels on interactive maps
• Pop-up labels, or labels only after click
• During zooming in, remove labels when their
feature goes outside the frame
• During zooming in, allow more and more labels;
during zooming out, remove labels of less
important features
Size of labels does not change!
Commercial packages
• Maplex for ESRI’s ArcInfo and ArcGIS
• EverName for MapInfo
• MapText for ArcInfo, ArcGIS, MapInfo,
GeoMedia, …
User Interface of EverName
Output of EverName
Output of MapText