Global approaches and local strategies for phase unwrapping

Phase unwrapping, i.e. the retrieval of absolute phases from wrapped, noisy measures, is a tough problem because of the presence of rotational inconsi stencies (residues), randomly generated by noise and undersampling on the p rincipal phase gradient field. These inconsistencies prevent the recovery o f the absolute phase field by direct integration of the wrapped gradients. In this paper we examine the relative merit of known global approaches and then we present evidence that our approach based on "stochastic annealing" can recover the true phase field also in noisy areas with severe undersampl ing, where other methods fail. Then, some experiments with local approaches are presented. A fast neural filter has been trained to eliminate close re sidue couples by joining them in a way which takes into account the local p hase information. Performances are about 60-70% of the residues. Finally, o ther experiments have been aimed at designing an automated method for the d etermination of weight matrices to use in conjunction with local phase unwr apping algorithms. The method, tested with the minimum cost Bow algorithm, gives good performances over both simulated and real data.