An efficient contextual classifier for land-use classification

Abstract

A new contextual classifier has been developed for information extraction from remotely sensed imagery. The algorithm is computationally very efficient, and experiment indicated that it can achieve more accurate results than the conventional maximum likelihood classifier and a number of commonly-used texture/contextual algorithms. The new contextual classifier includes two basic procedures: greylevel vector reduction and frequency-based classification. In greylevel vector reduction, the number of gray-level vectors in multispectral space was reduced using a new data-reduction algorithm through rotating multispectral space into eigen space. As a result, the multispectral data were reduced to images of one feature dimension with the loss of relatively little discriminant information. Each gray-level vector-reduced image was then used in the frequency-based procedure to derive useful information. The frequency-based classification procedure includes a grey-level vector occurrence-frequency extractor, a minimum distance classifier and an accuracy evaluator.