Establishing spatially continuous patterns in a non-supervised way

In the authors´ previous work (1998), the advantages of a continuous pattern approach versus a discrete pattern (homogeneous zones) were shown. In that paper, the training fields were introduced in a supervised way. This paper carries this research line further, developing a non-supervised version. Their aim is to extract the spectral classes and their spatial pattern. The test bed consisted of images acquired by the Thematic Mapper (TM) sensor on board the Landsat V satellite