An Approach to Building Extraction in Natural Image

A new approach to building extraction is proposed. Firstly, RPCL algorithm is used to improve segmentation granularity of super-pixel algorithm; Secondly, the spatial envelope pattern is used to classify image patches using support vector machine classifier, and distinguish building from non-building category in the image; Finally, building hypothesis in the image is verified based on parallel feature of the building in the image. The experiments on standard dataset show that the proposed algorithm outperforms other building extraction algorithms, and also meet the real-time requirements in general application.