A level set method for Bayesian tomographic reconstruction

We present a level-set method for Bayesian tomographic reconstruction. A novel image prior is derived from the mean curvature evolution of level sets of an image. As it has been studied in image processing with nonlinear diffusion, this prior encourages the stabilization of an edge while the reconstructed image is smoothed along both sides of the edge. An algorithm of iterated coordinate decent was implemented with the proposed prior using Brent´s method for one-dimensional optimization. Our simulation results demonstrated that our algorithm can outperform existing priors for preserving edges during tomographic reconstruction without introducing additional artifacts.