| Abstrakt: |
An algorithm for the segmentation of multispectral images is presented. First the algorithm performs a preliminary estimate of the number of segments, then the image is divided into segments using a noniterative version of the k-means method. Next, statistical analysis is performed; segment pairs that are realizations of the same random vector are found and merged. The results of testing the algorithm on model and real (HYPERION sensor) data are presented. Partitions of real images, whose segments correspond to various elements of the landscape, are shown. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink