Map Classification according to their Visual Global
Properties
Catherine Dominguès*, Tahar Kharchi**
* IGN, COGIT Lab., 73 avenue de Paris, F-94160 Saint-Mandé
** UPV – IRD, GRED Lab., 911, avenue Agropolis, F-34394 Montpellier
Abstract. Making on demand maps is a problem which is shared by plenty of national
mapping agencies. One strategy to help a map maker to make his/her map is based on
analogy. This strategy requires a large example base. Thus, the point is to explore this map
base in order to offer the map maker a reasonable number of relevant examples. In this
paper, a survey which aims to: i) evaluate the visual global properties of topographic map
examples and ii) make groups of maps which will be offered to mapmakers in the software
interface is presented. The results i.e. the map groups qualified by their visual properties are
shown.
Keywords: on-demand map, map visual property, survey.
1. Introduction
Making on demand maps is a problem which is shared by plenty of national mapping
agencies or cartographic companies. Indeed, the French national mapping agency (IGN)
offers a web service Carte à la carte (On-demand map) by which each Internet user can
specify the scale, the orientation, the title, the illustration, the center of the map to obtain
his/her own customized map.
A map can be seen as the result of the application of a legend to a geographic data set.
Different strategies may be used to help a map-maker to make his/her map. For example,
extracting color palettes from paintings and making different legend proposals from these
palettes (Christophe & Ruas 2009). Another strategy1 is based on analogy: instead of
acquiring cartographic knowledge, the map-maker can select maps among different map
examples. Secondly, the corresponding legends are applied to the map-maker's data to obtain
maps which are supposed to be like the original examples. This strategy requires a large
example base in order to give choices to the map-maker to better meet his/her needs.
Nevertheless, the larger the example data base, the more difficult it is to show all the
examples at the same time. Thus, the point of this proposal is to explore this map base in
order to offer the map-maker a reasonable number of relevant examples.
Previous work (Dominguès 2008) defines map visual global properties, some of which
are:<property>: realistic, precise, original, balanced. These properties may provide a way to
structure the example data base by defining map groups based on them. The software
interface helps the map-maker by guiding him/her towards relevant map groups according to
their properties. In this paper, the survey which aims to evaluate the visual global properties
1 This is the strategy which has been implemented in the IGN Carte à la carte.
of topographic map examples is shown in part 2. The statistical analysis of the survey allows
us to identify the most exaggerated maps according to each visual property; this is presented
in part 3. Results are discussed in part 4 before conclusions and future perspectives.
2. The survey
The survey aims to evaluate the properties of topographic map examples and to make groups
of maps based on these properties. The ability to have a higher response rate is one challenge
of this survey: the more people participating, the more reliable the survey data are. Thus it
was decided to use the World Wide Web to collect data regarding the appearance of the
maps.
The advantages of an online survey include rapid access to numerous potential respondents
and previously hidden populations, instant data collection (results are automatically stored in
a database), immediate feedback, lower cost of construction and postage. The most important
point is the similarity of experimental conditions, i.e. respondents are near or in the same
conditions when commenting on maps (object of the survey) or when constructing them by
use of a software, the maps are shown on the screen with varied and uncontrolled context
conditions.
Despite everything, online survey displays some disadvantages. The limitation of this
technology in this case can be summed up as: misunderstanding some questions by people
and lack of seriousness when answering them. Also, map appearance depends on the screen
size as well as on the screen calibration adjusted to each respondent.
2.1. Map description
Database maps are supposed to be examples of symbolization for map-makers. Indeed, the
map-maker cannot imagine the visual properties of his/her map by merely looking at the
map legend. Neither can he/she imagine them by looking at the application of the legend to
example data. Thus, in order to minimize the discrepancies between the visual properties of
the database examples and the map which is obtained by applying the example legend to the
map-maker's data, map examples are classified according to the landscape they represent.
The landscape is supposed to be described by two parameters: density and relief. Density
may be: urban or rural and relief: mountainous, plane or sea area. Consequently, the mapmaker
is supposed to know which landscape is predominant in his/her map to visualize
relevant examples. For instance, Figure 1 and Figure 2 show map samples with two different
landscapes, respectively: rural and mountainous area and urban and sea area.
Figure 1. Faroe Islandsdensity:
rural, relief: mountainous area.
Figure 2.Ukraine,
density: urban, relief: sea area.
Legends have been made in the course of previous work. Several are deduced from the
national mapping agencies' legends (Jolivet 2009): for example, the legend of the map in
Figure 1 is a transposition of the Faroe Islands to IGN data. Others are made into colors
extracted from qualified color palettes (Dominguès & Bucher 2006): for example, Figure 3
and Figure 4 show maps whose legend colors are respectively extracted from the calm and
the soft color palette.
Figure 3. Legend with calm color palette density:
rural, relief: mountainous area.
Figure 4. Legend with soft color palette density:
rural, relief: plane area.
Lastly, 48 maps have been selected. They have been made from 16 color palettes applied to
3 types of landscapes: rural and plane area, urban and sea area and rural and mountainous
area. All of them are topographic, based on IGN data.
2.2. Survey characteristics
The construction of the questionnaire can be illustrated as follows:
 48 topographic maps;
 one closed question about 4 properties repeated for every map: Do you think this map
is Realistic? Precise? Original? Balanced? …;
 6 response choices were given for each property: don't agree at all, disagree, neither
agree nor disagree, agree, strongly agree and no opinion. (see Figure 5, screen
capture of a page of the survey web site);
 property definitions in footnote page:
‐ Realistic map: does the symbolization chosen refer to reality?
‐ Precise map: does the map give the right information about the situation on the
ground at first glance?
‐ Original map: is the map distinguished from the habitual mapping uses?
‐ Balanced map: are the different elements in the map represented in a way that
none predominate the others?
 and the last page of the questionnaire permits to gather information about job, sex of
the respondents and their comments (see Figure 6: screen capture of last page of the
survey web site, is an illustration).
Moreover, a web site was designed. The online survey tools available do not permit such a
high number of pictures (48 maps) and their databases are generally limited to 100
responses, beyond which the survey tool shifts automatically to its paid version. So, the
alternative is to construct our own web site and connect it to a database. Firstly, it is free;
secondly, the administrator is allowed to provide his personalized graphical user interface
(GUI) adapted to the questionnaire features. Besides the design look, the database in which
responses, comments and all information about the questionnaire are stored, can easily be
organized which constitutes a reusable information database.
Technically, the web site was developed with:
 HTML (Hyper Text Markup Language): this is a web markup language used to create
web pages that can be displayed in a browser (or web navigator).HTML is used to
present content (text, image) in a webpage. It provides the tools with which the
content of a document can be structured or annotated according to different types of
metadata and indications of rendering and display;
 CSS (Cascading Style Sheets): this is a standardized syntax that gives web designers
extensive control over the presentation of their web pages and is an essential
component of web design;
 PHP (Hypertext Preprocessor): this is one of the most famous languages widely used
to create dynamic and interactive websites. PHP is generally associated to database in
order to facilitate storing of data stream they receive;
 MySQL is a RDBMS (Relational Database Management System). Its role is to store
data in an organized way to easily handle them later. Thanks to MySQL, responses of
surveyed people, their job and even their comments will be treated and requested.
Language used in this field is called SQL (Structured Query Language): It is standard
language designed for managing data held in relational database management
systems (RDBMS).
Finally, the website has been designed and tested locally via the server WampServeur to
verify receipt of responses in the survey database. The site was hosted by FranceServ, an
Internet hosting service company that allows individuals and organizations to make their
website accessible via the World Wide Web.
The survey web site was sent by email to all our contacts, inside and outside IGN
(Geographical National Institute). The idea was to reach different profiles, experts and nonexperts
in map-making.
The survey was also posted in GIS and mapping forums such as GeoRezo2, GIS Forum3,
Plaisance Practice4, Geomatics-Enit5. It was available from 27th June to 11th July 2012. Not
only is this strategy useful to reach a huge number of people with different profiles, but it
provides discussion topics around the survey as well.
2http://georezo.net/forum/viewtopic.php?id=80179
3http://www.forumsig.org/archive/index.php/t-34622.html
4http://www.plaisance-pratique.com/
5http://www.sitealytics.com/geomatics-enit.com/
Figure 5. Screen capture of a page of the survey web site.
Figure 6. Screen capture of last page of the survey web site.
2.3. Statistical analysis
Lastly, the survey database contains more than 270 responses. Respondent profiles are very
varied (see Figure 7). It is sufficient to carry out a statistical analysis for each property. The
purpose is to make map groups based on their visual properties and to identify the maps
which are the most <property> or the least <property>, in other words, maps which may be
a caricature of a visual property. For each property, the statistical analysis (Bressoux 2010)
allows us to identify the most exaggerated maps i.e. maps whose mark is the most different
from the average according to the standard deviation.
Figure 7. Respondent profiles.
The ordered qualitative choices have been translated into quantitative values to calculate
average and standard deviation, for each map and each property. So, firstly, for each map and
each property, the mark could have been calculated (the average of all the marks given in the
survey). Secondly, for each property, the average and the standard deviation have been
calculated (see Table 1).
Average
Standard
deviation
Realistic -0,10 0,45
Precise -0,14 0,39
Original 0,09 0,22
Balanced -0,35 0,47
Table 1. Average and standard deviation for each visual property.
The graph below (see Figure 9) shows value distribution of the realistic variable. Two groups
of maps can be distinguished at the two extremities: the most realistic and the least realistic
maps.
We can read on the positive x-axis: the more distant from the average the realistic value is,
the more realistic the map is. In the negative part: the more distant from the average the
realistic value is, the least realistic the map is.
Indeed, maps with values greater than the standard deviation are very realistic. On the
contrary, in the negative part, maps with values less than the value of the average minus the
standard deviation, are considered unrealistic.
Figure 9.Value distribution graph of the realistic property.
Therefore, the two groups related to the realistic property can be sorted. The examples below
give an idea about the two most realistic maps (Figures 10 and 11) and the two least realistic
maps (Figures 12 and 13):
Two most realistic maps
Figure 10. Germany legend density:
urban, relief: sea area.
Figure 11. Legend with calm color palette density:
rural, relief: mountainous area.
Two least realistic maps
Figure 12. UK legend density:
rural, relief: mountainous area.
Figure 13. UK legend density:
rural, relief: plane area.
3. Results and discussion
The statistical classification approach, in Figure 7, has been generalized by the other
properties: precise, original, and balanced. Eight groups of maps are constructed; they are
summed up in Table 2:
Map property Intensity Legend
Realistic Most
Germany/urban/sea area
Calm palette/mountainous area
Calm/rural/plane area
Calm palette/urban/sea area
Estonia/urban/sea area
Least
UK/mountainous area
UK/rural/plane area
Soft palette/mountainous area
Traditional/mountainous area
Traditional palette/rural/plane area
Tropical palette/mountainous area.
Precise
Most
Germany/urban/sea area
Calm palette/mountainous area
Calm palette/urban/sea area
Estonia/urban/sea area.
Least
UK/mountainous area
UK/rural/plane area
UK/urban/sea area
Traditional palette/mountainous area
Traditional palette/rural/plane area
Tropical palette/mountainous area
Tropical palette/rural/plane area
Original
Most
Earthy palette/mountainous area
Earthy palette/urban/sea area
Powerful palette/mountainous area
Powerful palette/rural/plane area
Powerful palette/urban/sea area
Regal palette/urban/sea area
Rich palette/urban/sea area
Soft palette/urban/sea area
Traditional palette/urban/sea area
Tropical palette/urban/sea area
Ukraine/urban/sea area
Least
Germany/mountainous area
Germany/rural/plane area
Germany/urban/sea area
Estonia/rural/plane area
UK/rural/plane area
Balanced
Most Estonia/urban/sea area
Least
Faroe/mountainous area
Faroe/rural/plane area
UK/mountainous area
UK/rural/plane area
Traditional palette/mountainous area
Traditional palette/rural/plane area
Table 2. Map groups.
According to this classification some remarks can be given:
 many maps are qualified very original, i.e. those maps are designed differently from
the usual mapping uses;
 few maps are very balanced. We suppose that some colors predominate among others;
 the color palette calm generates very realistic maps.
Furthermore, these map groups can be offered to mapmakers in Geoxygene: a GIS/map
making software by IGN.
The translation of ordered qualitative variables in quantitative variables is possible because
the study only considers one property at a time and can only be based on average and
standard deviation. Nevertheless, it would have been more correct to design the test answers
as a mark on a scale of 5 or more than as a qualitative answer among 5 ordered qualitative
answers.
4. Conclusions and future perspectives
The survey answers have been organized into a database. The latter constitutes a source of
information which is reusable for other work. In fact, additional variables such as age, sex,
job, may be used to refine the result survey.
This work is the first step to analogy-assisted map making. The next step is in progress. It will
consist in deducing global visual properties from elementary salient features. A semiotic
study (Landragin 2006) according to type of landscape is being performed. It aims to
highlight the map salient features, the elements that influence the choice, of people being
surveyed, relative to properties. This analysis will answer the question: how can we map out
our data in order to get which <property> map?
References
Bressoux P (2010) Modélisation statistique appliquée aux sciences sociales. Bruxelles : De Boeck
(2ème éd.)
Christophe S, Ruas A (2009) A process to design creative legend on-demand, 24th International
Cartographic Conference.
Dominguès C. (2008) Description de cartes géographiques. 8èmes journées francophones "Extraction
et Gestion des Connaissances" - Workshop on "User modeling and Web interface customization",
p. 59-68.
Dominguès C, Bucher B.(2006) Application d'aide à la conception de légende. Colloque International
de Géomatique et d’Analyse Spatiale SAGEO'06. Strasbourg (France)
Jolivet L.(2009) Characterizing maps to improve on-demand cartography - the example of European
topographic maps., 17th Conference on GIScience and Research in UK (GISRUK'09), poster session.
Durham (UK)
Landragin F. (2006) Visual Perception, Language and Gesture: A Model for their Understanding in
Multimodal Dialogue Systems. Signal Processing 86 (12), Elsevier, Amsterdam, The Netherlands,
p.3578-3595.