Datenbestand vom 25. Mai 2020
Tel: 089 / 66060798
Mo - Fr, 9 - 12 Uhr
Fax: 089 / 66060799
aktualisiert am 25. Mai 2020
978-3-8439-1445-1, Reihe Mathematik
Computational Aspects of Compressed Sensing
251 Seiten, Dissertation Technische Universität Darmstadt (2013), Softcover, B5
A central task in the novel field of Compressed Sensing is sparse recovery, i.e., finding the sparsest exact or approximate solution to an underdetermined linear equation system. Over the past decade, several conditions have been identified under which this generally intractable problem can be solved efficiently.
In this thesis, we investigate the computational complexity of ubiquitous sparse recovery conditions based on the spark, the restricted isometry property and the nullspace property of the sensing matrix. Furthermore, we provide an extensive numerical comparison of various state-of-the-art solvers for the most popular sparse recovery approach, the Basis Pursuit (l1-norm minimization) model, and demonstrate how a novel heuristic optimality check can significantly improve both speed and accuracy of most solvers. In addition, we introduce a new subgradient algorithm involving adaptive approximate projections for general nonsmooth convex constrained optimization and discuss its application to Basis Pursuit and variants.