Nonparametric classification methods are often useful in discriminating features or substances even when the functional term of the underlying distributions are unknown to the analyst. One such case is that of geological features, largely devoid of vegetation. Basically, nonparametric classification assumes that there exists a set of discriminant functions (one for each signature) with known functional form except for a set of parameters or weights. In this paper, a nonparametric classifier based on a least-square-error criterion is introduced. Using the designated training samples, an iterative procedure can be formulated which learns the values of the unknown parameters. Consequently, the classification problem is solved by computing the discriminant function and selecting the maximum. Example classifications of LANDSAT MSS scene are studied. Experimental results in the form of thematic maps and percent of correct classification are compared with other well-known techniques such as Bayes and density-slice methods.
Date of this Version