[Joint work with Gero Friesecke] We consider phase retrieval on the space of square integrable functions. Assuming object space knowledge of the image (such as positivity or support), we show that the Error-Reduction algorithm may be viewed as a discretized gradient flow (without the need to explicitly impose object space constraints). We use this setting to analyze stagnation properties of the Error-Reduction algorithm and propose a novel Error-Reduction variant that outperforms standard algorithms.

This presentation is part of Contributed Presentation “CP8 - Contributed session 8”

Authors:

Tsipenyuk Arseniy (Technical University of Munich)

Gero Friesecke (Technical University of Munich)

Keywords:

alternate projections, gradient flow, image reconstruction, inverse problems, nonlinear optimization, phase retrieval