Separating Multimodal Histogram to Unimodal Clusters

V. Sidorova (Russia)

Keywords

Clustering, multidimensional histogram, image analysis.

Abstract

The fast nonparametric technique of unsupervised classification of the multispectral data based on the construction of the multidimensional histogram and it's separating into unimodal clusters is considered. The technique does not use any suppositions about functions of distributions of various classes. It also does not require any priori data (number of clusters, number of iterations etc.). The system of hashing for a fast access to histogram is suggested. The technique was tested on three and five spectral space images of the Earth surface.

Important Links:



Go Back