OBJECT-BASED IMAGE ANALYSIS IN REMOTE SENSING APPLICATIONS USING VARIOUS SEGMENTATION TECHNIQUES

Extensive education, research and development activity is car- ried out in geoinformatics at Eotvos Lorand University (ELTE), Faculty of Informatics, in cooperation with the Institute of Geodesy, Cartography and Remote Sensing (F¨ OMI). It includes the teaching of subject "Remote Sensing Image Analysis", research of segment-based classification of remote sensing images and its applications in operational projects. Investigation of segmentation methods is embedded into the classification problem. Segments are homogeneous areas of images, consisting of neigh- boring pixels. Segment membership of pixels conveys valuable geometric information to classification step. This article gives a summary on several merge-based and cut-based segmentation methods. The application of segmentation is not only an option, but a necessity in the processing of very high resolution images, as their pixels usually cannot be interpreted individually. Segments are assigned with several attributes (e.g. texture) derived from geometrical properties. This leads to the advanced approach called Object-based Image Analysis (OBIA). As an application, the task of delimiting tree groups and scattered trees in pastures will be presented in detail. Three further applications will also be shortly introduced.