A three-dimensional feature space iterative clustering method for multi-spectral image classification

A practical method of three-dimensional feature space iterative clustering (3D-FSIC) for image classification has been introduced, in which the clustering iteration is performed in three-dimensional feature space rather than scanning the image pixel by pixel. This method permits the cluster size and pixel frequency to be taken into account so that a more advanced decision rule, the optimal multiple point reassignment (OMPR) can be applied. The paper also provides a simple technique for splitting a cluster based on the first principal component without performing principal component transformation. Finally, a classification example using hue images as well as a discussion of the advantages of using hue images in the 3D-FSIC classification is given.