MORPHOLOGICAL AND OBJECT-ORIENTED METHODS IN CLASSIFICATION OF MULTISPECTRAL DATA

Abstract. One of the preliminary tasks in supervised classification of multispectral data is the segmentation process. As result the geometry of distinct land cover classes under study is found. In this study we compared the applicability of two methods for image processing for information extraction from satellite images- namely object-oriented approach and mathematical morphological method. Mathematical morphology is a geometric approach in image processing and analysis with a strong mathematical favor. Originally, it was developed as a powerful tool for shape analysis in binary and, later, satellite images. The applications of the mathematical morphological methods could be summarized as follows: satellite images segmentation; use of morphological filters for generation of digital elevation models (DEMs); satellite images classification. The second method considered is the multivariate segmentation realized by the eCognition package. This patented algorithm is used as a reference in the comparison with the newly developed ones. Both methods were applied for segmentation of a satellite image of central Bulgarian region. The obtained results were compared and the advantages and disadvantages of morphological method are discussed. Conclusions about applicability of morphological methods for segmentation of satellite images are als