Most optimization methods popular in phase retrieval rely on gradients or proximal operators embedding phase less data onto an image space (and an illumination for blind ptychography), however convergence rate of these methods scales poorly with data size. By analyzing the relationship network and minimizing the difference between all pairs of overlapping frames, it is possible to identify the slowly-evolving principal modes, and solve large scale position drifts, or slow phase variations. Examples applications will be discussed.
Copyright. All Rights Reserved.