Matematiktorvet, Building 303B, Room 010
2800 Kgs. Lyngby
PhD Project: Training Sets in Large-Scale Reconstruction Methods
The project is affiliated with the
project High-Definition Tomography (HD-Tomo) financed by the
European Research Council.
This PhD project aims at providing a theoretical and methodological framework for the use of training sets as non-parametric priors for the solutions in tomographic image reconstruction. In particular we will study and further develop the use of artificially generated training images for incorporation of structural information about the solution.
Training images are conceptual images that contain a set of desired visual features, and we use dictionary learning to construct a dictionary from these images. Using the dictionary as prior information, we solve a Lasso least-squares problem that yields a reconstruction that has a sparse representation with respect to the dictionary.
Place of employment