A Simple and Extendable Segmentation Method for Multi-Polarisation SAR Images

Abstract

This work presents a simple feature-based multi-channel SAR segmentation method that produces good, smooth, fast and robust results for image segmentation and interpretation. A basic set of six features produce good class distinction for general applications, and the method is easily extendable with new features and for multi-source data fusion. Optional steps of sub-sampling and Markov Random Field based contextual smoothing are presented. The different stages of the basic approach are demonstrated with real data examples from several sensors and for several applications and is fast enough for operational use.