Robust quadrature filters

摘要: We deal with the relation between two well-known topics in signal processing and computational vision: quadrature filters (QF’s) Bayesian estimation Markov random fields (MRF’s) as prior models. present a new class of complex-valued MRF models such that optimal estimators obtained them correspond to output QF’s tuned at particular frequencies. It is shown machinery has proven be effective classical (real-valued) modeling may generalized complex case straightforward way. To illustrate power this technique, we implement robust exhibit good performance situations which ordinary linear, shift-invariant fail. These include are relatively insensitive edge effects missing data can reliably estimate local phase singularity neighborhoods; also for specification piecewise-smooth QF’s. Examples applications fringe pattern analysis, phase-based stereo reconstruction, texture segmentation presented well.