A fast and effective method of clustering of a set by distance to an edge set member is proposed in this paper. The method is appropriate for tasks of similarity search in multidimensional space as well as for traditional tasks of vector quantization. An example of such similarity search tasks is searching of similar block of the image in fractal compression or searching of images similar to given sample. A peculiarity of clustering for such task is presence only a nonlinear distance function between set members. Often this function depends from thousands of set members features. Effectiveness of the proposed method in comparison to well known median splitting method is shown for task of searching similar blocks of images.
clustering, vector quantization, image similarity, image lossy compression