High-Dimensional Grid-based Clustering for Multispectral Satellite Image Segmentation

The paper considers the problem of optimal data structure choice for grid-based clustering algorithms in high dimensional case. A special data structure is proposed that stores only non-empty cells avoiding large memory costs. The comparison results of the proposed method with the direct use of a hash table on multispectral satellite images are presented. They demonstrate a great advantage of the developed method in memory costs with comparable computation time.