Most of the maps contain wrong detection because of imperfections of the detectors or non-completeness of generic models of geographic features. So, we generate a new map using fusion to have a best credibility. For each type of feature, the rules describe the relevant fusion operators and parameters according to generic models.

The choice of fusion operators is important to obtain the desired behaviour as described by I. BLOCH [Bloch, 1994]:

• Conjunctive behaviour, reputed severe;
• Disjunctive behaviour, reputed indulgent;
• Intermediate, which performs compromises.

As linear features have same aspect on a given image, we use conjunctive fusion to have a best result. It removes wrong detection due to noise.

When a feature can be detected on several images, we use an intermediate fusion by computing the mean of the images of detection. Comparatively to disjunctive fusion it privileges conjunctive information regarding to disjunctive information, and does not cancel disjunctive information, as conjunctive fusion does.