ABSTRACT

This paper presents an Object oriented feature extraction approach in order to classify the linear features like drainage, roads etc. from high resolution Indian satellite imageries. It starts with the multiresolution segmentations of image objects for optimal separation and representation of image regions or objects. Fuzzy membership functions were defined for a selected set of image object parameters such as mean, ratio, shape index, area etc. for representation of required image objects. Experiment was carried out for both panchromatic (CARTOSAT-I) and multispectral (IRSP6 LISS IV) Indian satellite imageries. Experimental results show that the extraction of linear features can be achieved in a satisfactory level through proper segmentation and appropriate definition & representation of key parameters of image objects.