Dec. 19 2005 4:20PM

Dec. 19 2005 4:40PM

This paper presents an energy minimization approach to the deconvolution problem of multiple degraded frames of the original scene (single input multiple outputs). No prior information about the shape of degradation blurs is onsidered. The exact knowledge of the blur size is also not necessary and therefore any translation misregistration of input frames is automatically removed. The proposed regularization based on variational principles maintains stable performance even in the case of severe noise corruption. Further we demonstrate that the proposed technique can easily integrate resolution enhancement. Several experiments on synthetic and real data illustrate the robustness and usability of the proposed technique in real applications.

International Workshop on Data - Algorithms - Decision Making (Kontaktní osoba: Martin Janžura)