An improved hybrid clustering algorithm for natural scenes

Abstract

A new hybrid method for automatic clustering of satellite-observed natural scenes is presented. It uses a partitional clustering algorithm augmented by a hierarchical split-and-merge step at each iteration. The method also dynamically computes image-specific split-and-merge thresholds and can accommodate arbitrary information vectors. Better partitioning of the data and improved computational efficiency are achieved. The sensitivity of the method to the clustering parameters is presented, and the results show that the method is relatively insensitive to the choice of clustering parameters. Comparisons with the often used ISODATA algorithm show the effectiveness of the new approach.