The superposition of the signal from a blurred illumination results in partially coherent measurements. Here we propose the Gradient Decomposition of the Probe (GDP), a model that exploits the blurring kernel separability. We describe a first-order splitting algorithm GDP-ADMM to solve the .non-linear blind partially coherent phase retrieval problem. Remarkably, GDP-ADMM produces satisfactory results even when the ratio between kernel and beam size is more than one, or or when the distance between successive acquisitions is almost twice as large as the beam width.

This presentation is part of Contributed Presentation “CP9 - Contributed session 9”

Authors:

Stefano Marchesini (Lawrence Berkeley Nat'l Lab)

Huibin Chang (School of Math. Sci., Tianjin Normal University)

Pablo Enfedaque (Lawrence Berkeley Lab)

Hari Krishnan (Lawrence Berkeley Lab)

Keywords:

image reconstruction, inverse problems, phase retrieval