Building Recognition in Aerial Images

In this paper I propose the design of a novel building extraction system, entitled BuRS, which is capable of identifying man-made structures that are visible in monocular greyscale aerial photography of rural settlements (sparsely builtup areas). The system design is texture measure driven; integrates both edges and regions; and does not utilise an explicit model of the structures being sought for. One specific component of BuRS is investigated in detail, namely the region formation component. The region formation component incorporates a region growing operator. Two such operators have been implemented and tested the fractal error operator which is described in the literature and the homogeneous operator which is devised by the author. In comparative studies, it is shown that the homogeneous operator outperforms the fractal error operator for images of typically rural dwellings. An object-oriented approach is adopted for the design of BuRS’s software architecture and is discussed briefly.