Nonlinear Merge and Split Based Image Clustering Method Through Changing the Number of Clusters.

Nonlinear merge and split based image clustering method through changing the number of clusters is proposed. A nonlinear merge and split method for image clustering is proposed to overcome a complicated situation for the cluster’s distributions in a feature space which is relatively difficult to cluster with the linear approach like a k-means and ISODATA clustering algorithms. Through experiments, it is found that the proposed clustering method with the merge and split method by considering the within and between cluster variances, allows to overcome the complicated situation of the probability density function of clusters. It is also found that the proposed method is superior to the existing clustering method in terms of the similarity between the clustered resultant image and the topographic map of the corresponding map. Furthermore, the effectiveness of the proposed clustering method is shown with the real remote sensing satellite imagery data.