but every one can understand this share in problems. In other
words, the present SIG will not handle matter related to any
pure geometrical use of the data, e.g. image matching for geo-
referencing, or geo-referencing, accurate positioning,
assessment of digital terrain models, unless it appears
necessary for the wealth of our studies and if information
cannot be obtained from elsewhere.
All applications are foreseen, none is excluded. The already
foreseen work tasks are:
e list and better understand the fundamentals in data fusion
e list and better understand tools and methods in data fusion,
develop new ones
e develop and provide jnstruments for the assessment of the
quality of the fusion
e prepare application cases exhibiting several processes and
levels of fusion. These cases will help in illustrating data
fusion and in students training. Several domains (urban
domain, meteorology, ...) should be handled.
e prepare sets of data for well-documented sites which may
be useful for testing algorithms. This should be performed
with the help of the space agencies and other data providers.
3. THE NEED FOR TERMS OF REFERENCE
The concept of data fusion is easy to understand. However its
exact meaning varies from one scientist to another. Several
words have appeared, such as merging, combination, synergy,
integration, ... All of them appeal more or less to the same
concept but are however felt differently. There is also a fashion.
Several times, the word «fusion» is used while
« classification » would be more appropriate, given the contents
of the publication.
There is a need for terms of reference in the remote sensing
community. It has been strongly expressed in several meetings,
including those organised by EARSeL or SEE (see e.g., Van
Genderen, Pohl 1994; Wald 1997). The establishment of a
lexicon or terms of reference permits to the scientific
community to express the same ideas using the same words, and
also to disseminate their knowledge towards the industry and
'customers' communities. Moreover it is a sine qua non
condition to set up clearly the concept of data fusion and the
associated formal framework. Such a framework is mandatory
for a better understanding of data fusion fundamentals and of its
properties. It allows a better description and formalisation of
the potentials of synergy between the remote sensing data, and
accordingly, a better exploitation of these data.
The present communication aims at providing the basis for this
framework. It should be noted that this is not the only attempt
to set up definitions in data fusion. The remote sensing
community should not establish terms which are also used
elsewhere with different meanings. Therefore, whenever
possible, definitions were adopted which are already widely
used in the broad scientific community, especially that dealing
with information. Examples of such terms are image, features,
symbols, etc.
Several lexicons have been already set up. They have all been
established in the framework of the Defence domain (e.g., U.S.
Department of Defence 1991; DSTO 1994). Most of the terms
are part of the military jargon. They express needs of the
Defence which may be partly similar to those in other domains
where crisis occur, such as the management of a power plant.
However it is not easy to translate military terms in meaningful
words for the scientific community dealing with Earth
observation. Using these military lexicons would imply a
refinement of the military terms to expand their meaning, with a
reference to the time-space scales. It was concluded that using
an existing lexicon is not straightforward, and that a new one is
required to tackle the specific needs of our community.
However we should benefit from these previous works as much
as possible, and, whenever possible, we should use either the
terms already adopted or global architectures, etc.
The present communication summarises the discussions held
within the SIG since the first conference 'Fusion of Earth data'
held in Cannes, France, in February 1996. It proposes some
terms of reference which have met a consensus during the
second conference, held in January 1998.
4. THE EARSeL - SEE PROPOSAL
Data fusion means a very wide domain. It gathers a large
number of methods and mathematical tools, ranging from
spectral analysis to plausibility theory. Fusion is not specific to
a theme or an application. On the contrary the tools used in a
fusion process for a specific application may be tailored to that
case. It is very difficult to provide a precise definition of data
fusion. This large domain cannot be simply defined by
restricting it, for example, to specific wavelengths, or specific
acquisition means, or specific applications. Fusion process may
call upon so many different mathematical tools that it is
impossible to define fusion by these tools. For example, both
the simple sum of two images acquired by two different sensors,
and the more sophisticated encrustation of one image into the
other using the multiresolution analysis (Wald, Ranchin 1995),
are considered as fusion processes. Both implies at least a
preliminary geocoding of the data. A classification technique
based upon a sophisticated neural network is also a fusion
process.
Several definitions have already been proposed. They have been
discussed by e.g., Buchroithner (1998) or Wald (1997, 1998).
During the meetings of the SIG as well as in the conferences
'Fusion of Earth Data’, it was felt that most of these definitions
were focusing too much on methods though paying some
attention to quality. Some of them are restricted to sensors and
their output signals. As a whole, there is no reference to concept
in these definitions while the need for a conceptual framework
was clearly expressed in these meetings.
In data fusion, information may be of various nature: it ranges
from measurements to verbal reports. Some data cannot be
quantified; their accuracy and reliability may be difficult to
assess. In Earth observation domain, one may use some features
held in a geographical information system to help in classifying
multispectral images provided by several sensors. In this
particular case, some data are measurements of energy, and
others may be symbols.
Accordingly the definition for data fusion should not be
restricted to data output from sensors (signal). Opposite to most
of the published definitions, it should not be restricted to
methods and techniques or architectures of systems, since we
aim at setting up a conceptual framework for data fusion. Based
upon the work of Wald (1997, 1998), the following definition
was adopted: « data fusion is a formal framework in which are
expressed means and tools for the alliance of data originating
from different sources. It aims at obtaining information of
greater quality; the exact definition of ‘greater quality’ will
depend upon the application ».
This definition is clearly putting an emphasis on the framework
and on the fundamentals in remote sensing underlying data
fusion instead of on the tools and means themselves, as is done
usually. The latter have obviously strong importance but they
are only means not principles. Secondly it is putting also an
emphasis on the quality. This is certainly the aspect missing in
most of the literature about data fusion, but one of the most
delicate. Here quality has not a very specific meaning. It is a
generic word denoting that the resulting information is more
652 Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
sati.
pro:
incr
info
In t
con
diff
It |
com
COIT
defi
per!
are
inte
opp
bec
met
may
con
extr
Anc
opti
mez
anal
Kal
fore
Fus
leve
tern
foll
thec
Bije
Tou
Me:
call
of t
call
den
piec
rem
digi
cali
ang
con
An
mat
buil
exa
mul
be
hav
pix«
clas
pix
labe
wel
by |
time
vari
By
moi
obj:
eler
con
Son
