Scale hyperparameter estimation for GGMRF prior models with application to SPECT images

Abstract

In this work we develop a Bayesian reconstruction method for SPECT (single photon emission computed tomography) images, using as prior GGMRF (generalized Gaussian Markov random fields) distributions and estimating the scale hyperparameter following the evidence analysis. Preconditioning methods are used to estimate this hyperparameter and the approximations used are compared on synthetic images.