A detailed comparison between two fast approaches to urban extent extraction in VHR SAR images

Abstract

This work is devoted to he comparison of two algorithms for human settlement map extraction from VHR SAR data. The two approaches have been recently proposed in literature, but extensive comparison of their performance in different situation and in different areas of the world was not available yet. The first approach is based on the computation of local statistical indexes to detect "seed areas", which in turn are used to train a texture-based settlement detection procedure. The second approach is based directly on a textural feature, data range, and shows usually less precise, but equally useful results. In this work the two approaches are compared on a range of different SAR sensors, and a discussion of their relative performances for different spatial resolutions and radar frequencies is provided. Reference settlement extents are obtained from maps provided by global mapping projects.