Regularization of optic flow estimates by means of weighted vector median filtering

Vector median filtering has been recently proposed as an effective method t o refine estimated velocity fields, Here, the use of a weighted vector medi an filtering is suggested to improve the regularization of the optic flow f ield across motion boundaries. Information about the confidence of the esti mated pixel velocities is exploited for the choice of the filter weights. E xperimental results, on both synthetic and real-world sequences, show the e ffectiveness of the proposed procedure.