Performance analyses of probabilistic relaxation methods for land-cover classification

Abstract

The performances of two probabilistic relaxation (PR) classification methods, a standard and a modified version, are assessed in terms of classification accuracy measured by the Kappa coefficient and the CPU time required to carry out the computation. The classification results obtained with these methods are compared with results obtained using conventional maximum-likelihood classification (MLC). Experiments indicate that the modified PR method significantly improves upon the classification results generated by the MLC method. The modified PR method saves up to 70% of the CPU time, compared with the standard PR method, and also gives slightly better classification accuracy. © 1989.