Optimal recovery of 3D X-ray tomographic data via shearlet decomposition (Q360470)

Based on the 3D shearlet transform, the authors develop a new decomposition of the 3D X-ray transform. They derive a shearlet-based algorithm for the recovery of 3D images from X-ray data that are corrupted by additive Gaussian noise. For a given noise level \(\epsilon\) the involved thresholding scheme can be tuned to obtain an estimator that attains almost optimal mean square error rate \({\mathcal O}(\log(\epsilon^{{-1}}) \epsilon^{2/3})\) as \(\epsilon \to 0\). The results given on the mean square error rate rely on specific approximation properties of the 3D shearlets. They do not follow similarly as in the 2D case but require novel nontrivial technical tools.