ABSTRACT
Unsupervised classification is a process of grouping pixels that have similar spectral values. Each group of similar pixels is typically called a spectral class. There are many unsupervised procedures; in this chapter the people covers three general approaches: histogram-based unsupervised classification, sequential clustering and ISODATA clustering. The histogram from pixels of a uniform cover type is often bell shaped. Sequential clustering is a common strategy in unsupervised classification. The basic idea is to sequentially sample pixels and assign each sampled pixel to the nearest spectral class mean. There are three common ways to examine a classified image for spatial similarity of spectral classes: cursor inquiry, color palette manipulations, and overlay of spectral classes on the rectified image. It is important to look at the distribution of spectral classes across the entire classified image — not just in the areas they are familiar with.
