Automatic Building Extraction from VHR Satellite Image

Abstract

A satellite image contains natural region such as soil and vegetation and man created regions such as roads and buildings. Among these the precise location and identification of building features is one of the key information sources for urban planning, population estimation, land analysis, and environmental surveying. Previously low resolution satellite images were useful to analyze the features of buildings. Now-a- days due to the advancement of the technologies in the field of remote sensing, high spatial resolution imagery becomes available which provides more potential to automatically detect buildings. However, the high resolution image data contain rich information in the spatial domain which does not necessarily increase accuracy of the feature extraction significantly. Therefore, recent advances in the high resolution image processing is focused on the geometrical, spectral, statistical, contextual, and structural information extraction from the image. In this paper a technique to detect and to extract geometric features of the buildings in urban area from very high resolution (VHR) image has been proposed. The geometric features such as area, perimeter, centroid, solidity, convex area, are extracted. In order to analyze the performance of extraction, various edge detection mechanisms such as Sobel, Prewitt, Robert, Canny are implemented using Matlab. The performance analysis shows that canny operator with FCM clustering outperforms compared to conventional edge detection mechanisms.