Unsupervised Bayesian 3D reconstruction for non-destructive evaluation using gammagraphy

In this paper, a fully non-supervised extension of a recent reconstruction method is proposed. Previous contributions to Bayesian estimation of Gibbs hyperparameters are mostly based on the tabulation of the partition function. However, such a usual approach becomes impracticable when the dimensions of the object are large and variable across reconstructions. Our solution is to extrapolate the previously tabulated partition function to the specific dimensions of the region of interest. Validation tests show that this idea reveals efficient and realistic.