Bayesian 3-D tomographic reconstruction from limited numbers of radiographs

The authors consider 3-D tomographic reconstruction problems encountered using a small number of noisy radiographs. They present a Bayesian 3-D reconstruction method based on statistical models of the radiographic process and the generalized Markov random field (GGMRF) model for the 30D object. This model permits reconstruction of sharp density transitions in reconstructions. The authors present both the physical and probabilistic modeling issues and describe the techniques necessary to solve the optimization problems. They analytically present a technique for including Compton scattering in the reconstruction process and show the similarity in computation between the two cases