Image Processing Reference
In-Depth Information
of using the complementarity of the sources of information, in order to better identify
the limits of the image's homogenous components. For example, by fusing the pre-
cise graph provided by aerial imaging with the network's topological structure and its
approximate geometry provided by a map, we can come up with an excellent descrip-
tion of the road network.
Reconstruction. The multiplicity of points of view is an advantage for three-dimen-
sional reconstruction of observed scenes and, although this reconstruction may consist
of the typical methods (such as in stereovision, or in tomography), in other situations in
which acquisition is not controlled as well, it is only possible to reconstruct an approx-
imate three-dimensional information that empirically combines the various available
aspects.
Detection of change. This type of decision typically involves images taken at dif-
ferent dates, whether they are a map and an image, or multi-data images for tracking
crops or a pathology. It may also consist of sequences of multi-source images (at a
faster rate than multi-data images).
Updating knowledge of a phenomenon or a scene. Unlike in the previous case, the
decision consists here of using the information provided by different sources (possibly
multi-data) in order to modify or complete prior knowledge, for example, completing
a road network with new traffic circles to update a map.
Some of these different decision problems are similar to the combination of ex-
perts, since each image can be considered as an expert giving his opinion according
to his abilities. However, in general, the information in fusion problems of experts is
more scattered than with images. Learning is therefore more difficult because there is
less data, although the user often has less constraints over the algorithmic costs of the
methods. With images, the amount of data to be fused is both an advantage when it
comes to learning and a drawback for the computational load.
If we compare the problem of image fusion with that of data fusion based on
aggregation and multi-criteria optimization, we notice that one of the main differences
lies in the fact that for the latter, the goal is to find a solution that best satisfies a set
of generally stringent constraints, whereas in image processing, each source provides
(fairly explicitly) a level of satisfaction (for belonging to a category, for example,
which can then be considered a criterion) and the decision rather consists of choosing
the best one (the best category, for example).
3.2. Fusion situations
Depending on the applications, fusion problems can occur in different situations, in
which the types of information elements are not the same. The main fusion situations
in image processing are the following.
Search WWH ::




Custom Search