Glossary for Geospatial Science

maximum-likelihood classification:  A supervised image processing routine usually applied to multivariable images that have a dimensionality greater than 3.  First, the user selects sample areas, (called prototype or training set areas) for each feature to be mapped.  The maximum-likelihood process then computes the statistical properties of these features.  Similarly, it determines the statistical characteristics which separate feature types.  After this identification model has been built, the process tests each cell to determine in which prototype group it most probably belongs.  A threshold can be set so that if the highest probability of a feature match is below the threshold, the cell is designated as unknown.  This “catch-all” group insures that a cell representing a feature type not defined in one of the training sets will not be forced into membership.