We limited here our study to the assumption of a closed world, i.e. we suppose that a pixel must belong to one of the studied classes. An approach of opened world type, where it is possible to have an incomplete knowledge on the problem considered, would make it possible to suppose that a pixel can belong:

• either to one of the classes studied,
• or to another class not modelled in the application.

In the same way, in this first approach, we considered only singletons. However we already saw in the preceding pages that in an image, an element can belong to several classes. Taking this aspect into account, by allotting a mass of evidence to the union of all the possible classes, should make it possible to represent the reality in a better way.

In the implementation that we used here, we attributed a part of the initial mass of evidence to the representation of uncertainty. is the union of all the possible classes. This process allows Dempster-Shafer's rule, which has a conjunctive behaviour, to manage efficiently the conflict which can occur between the sources, and thus to fusion a large number of sources.

Dempster-Shafer's rule is close to the quantified adaptive fusion. It has the advantage of being associative, contrary to adaptive fusion. Moreover, its complexity is linear.

The advantages of Dempster-Shafer's rule are the following:

• a conjunctive behaviour;
• an efficient management of the conflict between sources;
• the possibility to fusion a large number of coherent sources;
• a linear complexity;
• associativity.

The main default of Dempster-Shafer's rule:

• the use of operators needing a long calculation time (multiplications).