Region-based change detection of PolSAR images using analytic information-theoretic divergence

In this paper a region-based change detection approach for multi-temporal PolSAR images is proposed. The PolSAR images are first segmented into compact local regions in the same way, then Wishart mixture models are learned to model each local region. To generate difference (DC) map, statistical distribution differences measured by information theoretic divergence are calculated for corresponding local region pairs. We adopt the Cauchy-Schwarz (CS) divergence as its analytic expression can be derived for Wishart mixture models. We test the proposed scheme on ALOS PALSAR PolSAR images. Qualitative and quantitative evaluations show its promising performance, compared to the traditional pixel-level approach.