A Bayesian Formulation of Coherent Point Drift

Coherent point drift is a well-known algorithm for solving set registration problems, i.e., finding corresponding points between shapes represented as sets. Despite its advantages over other state-of-the-art algorithms, theoretical and practical issues remain. Among issues, (1) it unknown whether the always converges, (2) meaning of parameters concerning motion coherence unclear. (3) relatively sensitive to target shape rotation, (4) acceleration restricted use Gaussian kernel. To overcome these provide different more general perspective algorithm, we formulate coherent in Bayesian setting. The formulation brings following consequences advances field: convergence guaranteed by variational inference; definition prior distribution provides basis interpretation parameters; rigid non-rigid can be performed single enhancing robustness against rotation. We also propose an scheme that applied non-Gaussian kernels greater efficiency than drift.