Urban area and building detection on high resolution multispectral satellite images using spatial statistics

With the increase in the resolution and the amount of satellite images, automatic extraction of urban areas and buildings became more important in the past decade. Extracting such information manually is tedious and needs a lot of expert effort. In this work, a system for detecting the urban areas, then finding the buildings inside these areas is proposed. LISA analysis is used for detection of urban areas. After the urban area is detected, a mean-shift based segmentation is applied; then each segment is decided as building or not by using segment-test on spectral features and Local Moran´s I value. Classification of buildings is done by KNN (K-Nearest Neighbor) classifier and Parzen classifiers. Input images to be used are 3 band multispectral images.