Using the l1-norm for Image-based tomographic reconstruction

This paper introduces an l(1)-norm model based on Total Variation Minimization for tomographic reconstruction. The reconstructions produced by the proposed model are more accurate than those obtained with classical reconstruction models based on the l(2)-norm. This model can be linearized and solved by linear programming techniques. Furthermore, the complementary slackness conditions can be exploited to reduce the dimension of the resulting formulation by removing unnecessary variables and constraints. Since the efficacy of the reduced formulation strongly depends on the quality of the dual-multipliers used when applying the reduction method, Lagrangian relaxation is used to obtain near-optimal multipliers. This allows solving larger instances in an efficient way.