Method for multi-spectral images segmentation based on the shape of the color clusters.

The paper describes an algorithm for multi-spectral images segmentation that takes into account the shape of the clusters formed by the pixels of the same object in the spectral space. The expected shape of the clusters is based on the Dichromatic reflection model(1), and it's extension(2) for optically homogeneous materials. Further the influence of the illumination and image formation by a color CCD camera are considered. Based on expected shape of clusters we propose a criterion of similarity/homogeneity for the extended region merging algorithm. This criterion works successfully in case of objects of voluntary shape and illumination by one or several sources of the same spectrum.