Efficient parameter estimation for discrete tomography using adaptive modeling

The recent increase in computational power has led to various advanced and powerful tomographic algorithms such as the Discrete Algebraic Reconstruction Technique (DART), where prior knowledge about the discrete grey levels is exploited to improve reconstructions. To achieve maximal reconstruction quality with DART, various configuration options are to be carefully chosen. As DART is an algorithm that requires a long time to compute, the estimation of the algorithm parameters generally is a daunting task. In this paper we, therefore, introduce a simple cost function to evaluate a chosen set of algorithm parameters. We also investigate the use of a state-of-the-art optimization technique, called adaptive modeling, where global optima of cost functions can be found with a limited number of costly function evaluations. Simulated experimental results show that using such an optimization technique leads to optimal reconstructions more quickly, without a penalty on their accuracy. (orig.)