Robust Motion Vector Relaxation for X-Ray Fluoroscopy Using Generalized Gauss-Markov Random Fields

We describe a Bayesian motion estimation algorithm which is part of a temporally recursive noise reduction filter for X-ray fluo-roscopy images. Our algorithm draws its robustness against high quan-tum noise levels from a statistical regularization, where a priori expecta-tions about the spatial and temporal smoothness of motion vector fields are modelled by generalized Gauss-Markov random fields. We show that by using generalized Gauss-Markov random fields both smoothness and motion edges can be captured, without the need to specify an often crit-ical edge detection threshold. Instead, our algorithm controls edges by a single parameter by means of which the regularization can be tuned from a median-filter like behaviour to a linear-filter like one.