Modified divisive clustering useful for quantitative analysis of remotely sensed data

A nonparametric, hierarchical, divisive clustering algorithm based on the mutual nearest neighborhood concept is proposed for classification of large volume of multispectral data indenting minimum memory and computational time. To realise the reduction of memory and computational time requirements, the authors define an area of interest around each pixel to search the required k nearest neighbors necessary for initiating splitting operation. The authors have proposed a modified distance measure, which incorporates the relative position of the pixels together with their feature values in order to improve the performance of the algorithm and reduce the classification error rate. The algorithm is simple, noniterative and can be successfully applied on any image data. Simulation experiments are conducted to corroborate the efficacy of the proposed methodology in terms of time, space and error rate.