Computational Imaging Based On Iterative Multi-image Phase Retrieval Techniques

2D coherent light field can be briefly illustrated by a pair of amplitude and phase functions.However,the phase measurement is acknowledged as a difficult problem because the frequency of visible light goes beyond the state-of-the-art detector's response frequency.Among the existing phase measurement techniques,iterative phase retrieval algorithms have attracted many researchers' attention due to its simple implementation and nonselective sample imaging.As the origin,single-shot phase retrieval algorithms suffer from twin-image problem,initialization sensitivity,oversampling requirement and so on.Thus,multi-image phase retrieval algorithms were designed afterwards.They have been applied to numerically reconstruct the complex object function and quite promising in the biomedical imaging.This paper will concentrate on the iterative multi-image phase retrieval technique and discuss its important issues.Based on the former work in our group,several new algorithms are put forward and their imaging characteristics are analyzed.Here is the summary of this paper.(1)As the two main components of multi-image phase retrieval technique,parametric optical transformation and computation structure are discussed through theoretical analysis and numerical simulation.Single-shot and multi-image strategies are compared from the perspective of time-frequency analysis.It is verified that the latter holds the advantages of initialization insensitivity,twin-image elimination,support-constraint waiver and sampling rate decrease.Besides,the serial and parallel computation structure are compared and given the explanation from the perspective of geometric analysis in optimization.This lays the foundation for later new algorithm design.(2)To guarantee the efficiency of multi-image phase retrieval algorithms in some special scenarios,like when the photon number is scarce,the noise effect should be considered.Thus,kinds of experimental noises are modeled and the existing multiimage phase retrieval algorithms are tested under different levels of them.A noiserobust multi-image phase retrieval algorithm is proposed.Its superiority is validated in two typical computational imaging systems.(3)The pixilation problem hinders the final imaging resolution of the lensfree imaging system.Thus it is firstly discussed and modeled.Then,by combination of pixel super-resolution algorithm based on interpolation and reconstruction and multiimage phase retrieval algorithm,a new noise-robust pixel super-resolution multiimage phase retrieval algorithm is proposed.The parameter effect on its convergence and reconstruction is analyzed.It is experimentally demonstrated that the new algorithm can break the pixel pitch barrier with a large field of view,thus achieving the high-throughput imaging with phase retrieval.(4)A computational 3D complex amplitude imaging is introduced.It incorporates the multi-slice model into the Fourier ptychographic microscopy.In the scheme,the light refocusing result is input as the initialization and can accelerate the algorithm convergence.The theoretical imaging resolution of the imaging modality is deduced from its 3D coherent transfer function,which will benefit the practical system design.Finally,an experiment on the Spirogyra algae sample demonstrates the computational imaging scheme.To conclude,this paper provides several new computational imaging strategies and solves the existing noise and pixilation problem.A 3D imaging modality is introduced.All these results will promote the application of iterative multi-image phase retrieval techniques.