Phase Retrieval Utilizing Particle Swarm Optimization

Phase retrieval is an important tool for image recovery techniques based on Fourier spectrum. Different iterative algorithms have been developed to retrieve phase information. However, due to the nonconvex feature of the phase optimization problem, it remains a challenge to globally obtain the optimal phase information. In this work, we proposed an iterative algorithm to retrieve the global optimal phase information by adopting particle swarm optimization technique to the hybrid input-output scheme. By escaping the local minima using stochastic perturbations and information exchange among particles' local solutions, the proposed scheme increases the possibility of reaching the global minimum in phase retrieval optimization. In the numerical simulations, the images reconstructed by the proposed scheme have an averaged mean-square error of 0.0055, which is, respectively, 43.88% and 36.78% smaller than those of the images reconstructed by hybrid input-output and guided hybrid input-output schemes. The feasibility of the proposed scheme was demonstrated by the results from actual experiments, which showed an agreement with the simulation. The proposed scheme is statistically capable of obtaining accurate phase information, and, therefore, can be applied to Fourier spectrum based image recovery techniques.