Multispectral images segmentation algorithm

Multispectral images segmentation algorithm is presented. First, preliminary estimation of segments number is obtained. Second, image is decomposed into segments by non-iterative version of the k-means method. Third, statistical analysis is carried out; couples of segments which are realizations of the same random vector are found and merged. The results of testing algorithm on model and real (sensor HYPERION) data are presented. Real images segmentations where segments correspond to the different elements of the landscape are given.