Evaluating unsupervised classifiers with similarity and comparison matrices

摘要

In this work, the evaluation of the unsupervised classification by the matrix with the measures of similarity is being presented. This matrix has been created with the classified images of the same Landsat MSS image but classified with different unsupervised classifiers. Three classification measures of similarity and two clustering methods for each similarity measure were used. For the evaluation of these classifiers a matrix with the measures of similarity has been estimated, which gives the pairwise relations between image clusters, after classification. The results indicated that the optimum classifier for the sample area was the one that used the measure of similarity, that is the Euclidian distance.