Edge-preserving Map Image Reconstruction From Projections: Gaussian And Poisson Data

Abstract

A new edge-preserving prior model, incorporating a line process, is proposed for maximum a posteriori tomographic reconstruction. This model allows trade-off between smoothness and edge-sharpness. Mean field ap proximation is used to obtain iterative aolutions for both Gaussian and Poisson data. In the later case, a new Expectation-Maximization type algorithm is derived. Reconstruction results for both cases are better than those obtained by conventional Filtered Back Projection and Maximum Likelihood methods. The new prior model and algorithms can be applied to sensor fusion and restoration of blurred images contaminated by Poisson or Gaussian noises.