PARALLEL ALGORITHMS FOR PHASE UNWRAPPING BASED ON MARKOV RANDOM-FIELDMODELS

A general framework is presented for the design of parallel algorithms for two-dimensional, path-independent phase unwrapping of locally inc onsistent, noisy principal-value phase fields that may contain regions of invalid information. This framework is based in Bayesian estimatio n theory with the use of Markov random field models to construct the p rior distribution, so that the solution to the unwrapping problem is c haracterized as the minimizer of a piecewise-quadratic functional. Thi s method allows one to design a variety of parallel algorithms with di fferent computational properties, which simultaneously perform the des ired path-independent unwrapping, interpolate over regions with invali d data, and reduce the noise. It is also shown how this approach may b e extended to the case of discontinuous phase fields, incorporating in formation from fringe patterns of different frequencies. (C) 1995 Opti cal Society of America