In imaging science, efficient acquisition of images by few samples with the possibility to precisely recover the complete image is a topic of significant interest. The area of compressed sensing, which in particular advocates random sampling strategies, has had already a tremendous impact on both theory and applications. The necessary requirement for such techniques to be applicable is the sparsity of the original data within some transform domain. Recovery is then achieved by, for instance, $\ell_1$ minimization. Various applications however do not allow complete freedom in the choice of the samples. Take Magnet Resonance Imaging (MRI) for example, which only provides access to Fourier samples. For this particular application, empirical results still showed superior performance of compressed sensing techniques.
\\
\\
In this talk, we focus on sparse sampling strategies under the constraint that only Fourier samples can be accessed. Since images -- and in particular images from MRI -- are governed by anisotropic features and shearlets do provide optimally sparse approximations of those, compactly supported shearlet systems will be our choice for the reconstruction procedure. Our sampling strategy then exploits a careful variable density sampling of the Fourier samples with $\ell_1$-analysis based reconstruction using shearlets. Our main result provides success guarantees and shows that this sampling and reconstruction strategy is optimal.
\\
\\
This is joint work with Wang-Q Lim (Technische Universit\"at Berlin).