| Summary: | Recent image classification based on watershed algorithm (Watson et. al.,1992) has reported as a potential method for producing accurate method of classifying satellite images. This method, however, can only be applied to 2-D feature space and hence several independen analysis had to be combined to give acceptable results. In this paper, a generalizat n of the watershed algorithm to enable classification of any n-D feature space is presented and analysed. Results are presented for 3 different models of classification: (i) maximum likelihood, (ii) combinations of 2-D watersheds, and (iii) n-D watershed, tested over a relatively heterogeneous land cover of Malaysia. It is shown that the n-D watershed method is the most superior, although it has not reported a substantial improvement in the classification accuracy from the 2-D approach. Given the elaborate computational time taken by the n-D watershed, it is concluded that the n-D watershed is no better than the 2-D approach.
|
|---|
