ICARE system is conceived for automatic cartography when the map required results from a classification, a vegetation map for example. So, the expert knowledge dealt with, is related to classification problems.

This means that the knowledge concerns the "classes" that we are looking for in the image. On the other hand, this knowledge handles information existing or computable on geocoded information (for example a geographic information system) or eventually on the image itself. This information may also be extracted from existing maps.

Expert knowledge is usually expressed in quite various ways, but we consider that expert knowledge is expressed in natural language, for example in the special case of vegetation types.

Let us take a theoretical class "PINES" for example and the corresponding expert knowledge being expressed in the following way:

In this sentence,

• "Principally" shows uncertainty about knowledge or a frequency degree.
• "SOUTH SLOPES FROM 800 TO 1500 M" concerns the D.E.M. (Digital Elevation Model).

This knowledge unit will be translated into the following production rule:

IF class "PINES" THEN     (south slopes) and (800m < elevation < 1500m)

is the certainty factor of the rule itself and evaluates the confidence in the rule:

• -1 you are sure it is NOT,
• +1 you are sure it is YES

In fact, it may be considered as the frequency degree for this particular class to be in the special context described by the combined conclusion of the rule. In the previous example, corresponds to "principally".