Quantified adaptive fusion such as we modified it behaves very well when the number of sources to be merged increases. The class having the largest consensus is always selected, and this system shows its efficiency when new sources are added. Generally, adding information leads to an improvement of classification.

For conjunctive fusion, the addition of out-image data to spectral bands leads to a great increase of the number of unclassified pixels. In effect, it is quite impossible to obtain the unanimity of the sources on the assignment of a pixel to a class.

On the other hand, quantified adaptive fusion manages the additional sources very well. The areas of classification obtained for each class are very homogeneous. All the classes are well recognized, except for the recognized class 7 with 58.17%, and to a lesser extent of the recognized class 8 with 71.90%.

We note on this example that, when the sources are numerous but globally concordant, a majority of sources in agreement is sufficient to decide to assign a pixel to a class. This process works well because the sources of information to be merged are globally coherent. Most of the time, we do not have a very severe conflict to solve. In this case, a fusion of conjunctive type is suitable. Unanimity is not essential, and at the opposite, it locks the decision-making process since it becomes difficult to obtain the unanimity when the number of sources increases.

Table 34: Quantified adaptive fusion of the spectral bands MSS4 to MSS7 and out-image data: distance to roads, plateaux, vales, slope orientation, slope steepness, edges, irrigation, DEM and distance to urban areas.

Rates of classification obtained
Class	Number of pixels   correctly classified (A)	Number of pixels in samples      (B)	Rate of pixels     correctly classified (A/B)
1	414	459	90.20%
2	437	459	95.21%
3	262	306	85.62%
4	339	391	86.70%
5	406	459	88.45%
6	401	459	87.36%
7	267	459	58.17%
8	330	459	71.90%
9	458	459	99.78%
TOTAL	3314	3910	84.76%