Treffer: A Bayesian Approach for Building Detection in Densely Build-Up High Resolution Satellite Image

In this paper, we present a novel automatic approach for building detection from high resolution satellite image with densely build-up buildings. Unlike the previous approaches which normally start with lines and junctions, our approach is based on regions. In our method, first the prior building model is constructed with texture and shape features from the training building set. Then, we over-segment the input image into many small atomic regions. Given the prior building model and the over-segmented image, we group these small atomic regions together to generate region groups which have a similar pattern with the prior building model. These region groups are called candidate building region groups(CBRGs). The CBRGs grouping and recognition problems are formulated into an unified Bayesian probabilistic framework. In this framework, the CBRGs grouping and recognition are accomplished simultaneously by a stochastic Markov Chain Monte Carlo(MCMC) mechanism. To fasten this simulation process, an improved Swendsen-Wang Cuts graph partition algorithm are used. After obtaining CBRGs, lines which have strong relationship with CBRGs are extracted. From these lines and the CBRG boundaries, 2-D rooftop boundary hypotheses are generated. Finally, some contextual and geometrical rules are used to verify these rooftop boundary hypotheses. Experimental results are shown on areas with hundreds of buildings.