Enhanced multiple-plane phase retrieval using evolutionary selection

Abstract

Phase retrieval techniques are necessary to overcome the ambiguities and non-uniqueness of the phase solution caused by the limitation of recording devices to capture phase information. Evolutionary algorithms are a family of population-based stochastic optimization algorithms that take inspiration from biological processes. Selection methods instigate selection pressure and determine how candidate solutions (individuals) are chosen among the population based on their fitness values. This study demonstrates the integration of ranking and roulette wheel selection methods in minimizing the amplitude mean squared error between intensity measurements to promote faster convergence in a multiple-plane phase retrieval setup. These selection methods are modified into the single-beam multiple-intensity reconstruction (SBMIR) technique, and assessed in terms of phase reconstructions and convergence graphs against the conventional algorithm.