MODIFIED CONTEXTUAL CLASSIFICATION TECHNIQUE FOR REMOTE-SENSING DATA

Conventional techniques of classification make use of the spectral inf ormation at each pixel to predict the class of that pixel independentl y of the observations at other pixels. Contextual techniques, on the o ther hand, utilize the information from other neighboring pixels also. Two methods of contextual classification exist - one for low-resoluti on data and one for high-resolution data. A new method is proposed by combining these two methods. The new method is compared with the Gauss ian maximum-likelihood classification and the two methods of contextua l classification for low-resolution and high-resolution data. Classifi cation algorithms are compared using normalized classification accurac ies and the Kappa statistics.