This talk presents a highly efficient Bayesian computation methodology to solve imaging inverse problems with unknown regularisation parameters. A main novelty of this methodology is that it uses stochastic optimisation algorithms that are driven by proximal Markov chain Monte Carlo samplers, tightly integrating modern high-dimensional Monte Carlo sampling and convex optimisation approaches. The proposed methodology is demonstrated on several challenging imaging problems and compared to other techniques to set regularisation parameters.

This presentation is part of Minisymposium “MS54 - Hybrid Approaches that Combine Deterministic and Statistical Regularization for Applied Inverse Problems (4 parts)”
organized by: Cristiana Sebu (University of Malta) , Taufiquar Khan (Clemson University) .

Authors:

Ana Fernandez Vidal (Heriot-Watt University)

Marcelo Pereyra (Harriott-Watt University)

Keywords: