Model-based iterative reconstruction for neutron laminography

Abstract

Neutron-based parallel-beam laminography is an important 3D characterization tool because it can image thick specimens with unique shapes and provides a complimentary contrast to X-rays for several elements relevant to the material sciences and biology. However, the inversion of neutron laminography data is complicated because of the non-traditional geometry of the set-up, the presence of noise and the occurrence of gamma hits on the detector during the course of an experiment. In this paper, we present a model-based/regularized-inversion reconstruction algorithm for neutron laminography. We introduce a new forward-model/data fitting term and combine it with a flexible regularizer function to formulate the reconstruction as minimizing a cost-function. We then present a novel optimization algorithm that is based on combining a majorization-minimization technique with a first-order method that is amenable to simple parallelization on multi-core architectures. Using simulated and experimental data, we demonstrate that it is possible to acquire high quality reconstructions compared to the typically used filtered-back projection algorithm and algebraic reconstruction techniques.