A Bayesian approach for SAR interferometric phase restoration

The phase unwrapping problem for Interferometric SAR (InSar) has been approached with a number of local and global methods, to cope with the intrinsic noise due to the SAR image generation. The authors propose a new filtering technique for interferometric phase images, based on a Bayesian approach, such that the subsequent unwrapping can be performed by a path independent integration. Assuming that the original phase profile has bounded differences, they introduce an a priori Gibbs distribution for the data that encourages images with conservative phase gradient. The resulting maximum a posteriori estimation can be recast as a maximum likelihood optimization with a zero-residue constraint. The simulations on synthetic data, carried out by a Metropolis version of the simulated annealing algorithm, show satisfactory results as the restored phase profiles are very accurate even for low values of the coherence.