Change Detection in High-Resolution SAR Images Based on Jensen–Shannon Divergence and Hierarchical Markov Model

Abstract

This paper addresses the problem of change detection in high-resolution multitemporal synthetic aperture radar (SAR) images. We propose to use Jensen-Shannon divergence (JSD) to measure the dissimilarity of the two scenes acquired at different times for deriving the difference map (DM). We figure out this divergence in a nonparametric way by introducing a direct density ratio estimation, making the DM generation free of distribution assumption. We also present a multiscale change detection framework which can capture and combine change cues at different scales. First, the coregistered SAR image pairs are decomposed into different scales by multiscale decimated wavelet transform (DWT). Next, the DM in each scale is generated by computing the local JSD. These DMs are then represented by a hierarchical Markovian model based on a quad-tree structure. The change map is finally inferred relying on hierarchical marginal posterior mode (HMPM). Experimental results on multitemporal TerraSAR-X images demonstrate the effectiveness of the proposed approach.