A Regularization Approach for InSAR and Optical Data Fusion

This paper investigates the joint use of interferometric SAR and optical data for 3D reconstruction. A framework for phase filtering constrained by the discontinuities of the optical image is presented. First, both the amplitude and the interferometric phase are projected in a 3D coordinate system. The problem is then expressed as the regularization of the amplitude and phase images with the introduction as prior knowledge of the edges detected on the optical image. We define the regularized elevation in the framework of Markov random fields (MRF) and derive a smoothness prior that both preserves sharp boundaries (based on total variation minimization) and is driven by the structures present in the optical image. We apply a recent graph-cut based algorithm to perform fast regularization of the elevation field. First results on a real pair of optical and InSAR images are presented.