Phase retrieval utilizing geometric average and stochastic perturbation.

• Investigate converging performance of existing phase retrieval algorithms. • Propose hybrid algorithm, faster to yield better results than existing algorithms. • Specifically, 40% faster to yield results with an averaged MSE 19.8% lower. • Address the problem of choosing between speed and quality in phase retrieval. Phase retrieval plays an important role in Fourier spectrum based image recovery techniques. Despite the fact that many algorithms have been proposed to improve the performance of the phase optimization, it is an inherent contradiction that in order to increase the possibility of reaching the global minimum by introducing more randomness during the searching, the optimization speed is jeopardized. In this work, we investigate the different performance properties of existing phase retrieval algorithms, and propose a Hybrid algorithm, which adaptively utilize geometric average or stochastic perturbation, to achieve fast phase retrieval with a high possibility of reaching the global minimum of the optimization. Simulation and experimental results show that the proposed Hybrid algorithm yields results with averaged MSE 19.8% smaller than that of the existing GHIO, and reaches low MESs 40% faster than the existing PSHIO does. With its feasibility demonstrated, the Hybrid algorithm can be used in Fourier spectrum based image recovery techniques. [ABSTRACT FROM AUTHOR]