A Bottom-Up/Top-Down Hybrid Algorithm for Model-Based Building Detection in Single Very High Resolution SAR Image

Abstract

Building detection from high-resolution synthetic aperture radar (SAR) image is an essential issue for many SAR applications in urban areas. In this letter, we propose a novel bottom-up/top-down hybrid algorithm for model-based building detection from single very high resolution (VHR) SAR image. First, the building model is generated and described by a set of extraction criteria, which restrict the spatial layout of a building and its primitive features. Specifically, the rectangles of different intensity levels are extracted from the SAR image as primitive features. Then the bottom-up stage proposes building candidates composed by extracted rectangles, and the top-down step predicts building candidates composed by weak features omitted in the primitive extraction. After that, all candidates are verified through false alarm detection. Under this framework, the detection performances can be greatly improved especially in dense built-up areas. The effectiveness of the proposed method is verified by experimental results obtained from real VHR SAR images.