Design And Implementation Of Multi-wavelength Phase Retrievalalgorithm

Phase retrieval refers to the restoration of the original signal by the amplitude of the signal Fourier transform or other linear transformation.Usually the observing device can only measure the intensity information of the signal,causing the phase to be lost,but the phase contains about 75% of the total image,so the phase of the restored image is particularly important in image processing.This paper mainly studies the application of phase retrieval in optical field from two aspects.Firstly,we study the use of lensless tunable lasers to tune out different wavelengths of laser light,record different wavelengths diffraction patterns as prioris,this method avoid moving aperture and illumination surface.The laser can be seen as an ideal light source.In the paper,the loss function is combined with the total variation of the regularization term,and multiple illumination wavelengths phase retrieval algorithm based on total variation is proposed to achieve phase retrieval by solving the minimization problem.The experimental results show that the proposed algorithm can effectively improve the quality of image reconstruction and robust for noise.Secondly,because the laser as the experimental light source is too expensive,and is too heavey to carry,in order to further simplify the optical system,we use low-cost light emitting diode as a light source.Since the spectral range of the light source is about 60 nm,only use the central wavelength will lost amount of information.In this paper,the spectral demultiplexing method is used to solve this problem.The broad spectrum is decomposed into multiple narrowband spectra.The number of narrowband spectra decomposed is more close to the actual spectrum of the diode.In this paper,based on the spectral demultiplexing,a new cost function is proposed by combining the full variation normal term,and the phase retrieval optimization problem under the condition of the ideal light source is solved by the alternating direction multiplier method.The experiments show that the quality of image reconstruction has been improved.Finally,since the phase retrieval problem is a nonconvex optimization problem,it can not be solved by using the convex optimization method or the direct derivative isequal to zero,which makes the calculation face great challenge.In this paper,based on the spectral demultiplexing,the majorization minimization algorithm is used to construct the convex agent function of the nonconvexian loss function.Then,we combine with the total variation regularization term.Combined with the support constraint,the gradient projection descent algorithms to realize image phase retrieval.Reflect the effectiveness of the algorithm through the experiments with many images.