Many image processing problems involve reconstructing an unknown in the face of incomplete data. The reconstruction is often formulated in terms of an ill-posed problem which is stabilized through regularized least squares.
Convergence properties of stabilized solutions and their discretizations are well known in many situations except when regularization vanishes on a fixed grid. The importance of accurate discretization in such cases will be motivated by examples.
It will also be demonstrated that standard methods fail to provide consistent discretizations in the limit of vanishing regularization, and alternative consistent methods will be presented along with a summary of analysis results.
Universität zu Lübeck
Ratzeburger Allee 160
23562 Lübeck
Tel. +49 451 3101 0
Fax +49 451 3101 1004
Universitätsklinikum
Schleswig-Holstein (UKSH)
Tel. +49 451 500 0
für die Ukraine