Accelerating ptychography with spectral initializations

Abstract

Combining synthetic aperture approaches with reference-less setups, ptychography is a promising phase retrieval technique for label-free quantitative phase imaging. Within the phase retrieval community, spectral methods are known to accelerate gradient descent schemes, however their positive effect on experimental ptychographic datasets has not been proved.\nInspired by the latest theories on optimal spectral estimation, we achieved 3 times faster ptychographic reconstructions than with a standard gradient descent algorithm, in both simulations and experiments. The algorithms and experimental parameters crucially impacting the convergence speed are discussed.\nWe believe that spectral methods will help improve both theoretical understanding and experimental implementations of ptychography.