ABSTRACT
This chapter shows how to use training data to generalize to a complete labeling, or thematic map, for an entire scene. The choice of training areas which adequately represent the spectral characteristics of each category is very important for supervised classification, as the quality of the training set has a profound effect on the validity of the result. Finding and verifying training areas can be laborious, since the analyst must select representative pixels for each of the classes by visual examination of the image and by information extraction from additional sources such as ground reference data, aerial photos or existing maps. Unlike supervised classification, unsupervised classification, or clustering as it is often called, requires no reference information at all. Instead, the attempt is made to find an underlying class structure automatically by organizing the data into groups sharing similar characteristics.
