The results for each scanner were analysed independently. Eight
predefined classes were used in each case for the unsupervised
classification algorithms considered in this work. In all of these
algorithms, the resulting map showed the existence of affected areas.
Post-analysis reduced the number of classes. The number of classes
decreased from 8 classes (the initially defined number of classes) to 4
classes (including the null class), which were those that had real
meaning (thematic classes). These classes were: 1) Clean ashlar
(without deposits or without moisture evidences), 2) lime/mortar
remains and 3) ashlars with a high moisture content. The values of the
statistical measures e overall accuracy and Kappa coefficient e were
obtained for each of the classification algorithms considered