Discontinuity-preserving vector smoothing of multivariate MR images using vector mean field annealing

We consider PD-, T1-, and T2-weighted magnetic resonance images jointly as a vector-valued image and use the angle of this vector field to formulate m aximum a posteriori restoration as a global optimization problem. We use Me an Field Annealing (MFA) to find restorations that are superior to those ob tained by previous multivariate approaches when shading artifacts near the MRI antenna are significant. Local homogeneity of the vector field as well as the angle between the components or the ratio of the components of the f ield are shown to have potential use for improving segmentation.