Research On Phase Retrieval Algorithm Of Single Coded Diffraction Pattern

The measured value in the coded diffraction pattern system is the amplitude data,which loses the phase part containing a lot of information.It is the phase retrieval problem to recover the original signal from the amplitude data.In classical phase retrieval algorithm,many coded diffraction patterns must be used to reconstruct the original signal accurately.However,it is not easy to reconstruct the coded diffraction patterns repeatedly in practical experiments.It is necessary to study how to reconstruct the original signal accurately only with a single coded diffraction pattern.In this paper,the phase retrieval scheme of a single coded diffraction pattern is studied.The specific work is as follows:Firstly,on the basis of Approximate Message Passing for Amplitude Based Optimization,a phase retrieval model suitable for coded diffraction pattern system is deduced.In this model,a denoising operator is introduced,and a phase retrieval algorithm based on denoising operator and AMP.A,which is suitable for single coded diffraction pattern,is proposed.In this algorithm,the denoising operator is used to replace the modified parameters in message passing.Experiments show that the algorithm can reconstruct high quality images using single coded diffraction pattern contaminated by Gauss noise.Secondly,in the convex optimization phase retrieval model,the initial estimation is required and the image cannot be reconstructed effectively under the condition of single coded diffraction pattern.In this paper,the regularized phase retrieval algorithm based on dual-tree complex wavelet transform is incorporated into the convex optimization phase retrieval model,and a phase retrieval algorithm based on convex optimization and dual-tree complex wavelet transform is proposed.In the process of convex optimization,the regularization algorithm is used to estimate the initial image signal,which can inherit the accuracy of the convex optimization algorithm and the effectiveness of the regularization algorithm in solving the single diffraction pattern problem.Experiments show that the algorithm can reconstruct the original image quickly and accurately from a single coded diffraction pattern,and has strong robustness.Finally,the image self-similarity is introduced into the phase retrieval model,and the non-local Bayesian estimation is used to establish the constraints of image self-similarity.A phase retrieval algorithm based on non-local Bayesian and image self-similarity is proposed.Then,some lost image information can be reconstructed by image denoising operator when solving inverse image problems.A phase retrieval algorithm based on non-local Bayesian and denoising operator is proposed.Experiments show that the phase retrieval algorithm based on non-local Bayesian estimation and image self-similarity can reconstruct the image accurately under the condition of a single coded diffraction pattern,and the phase retrieval algorithm based on non-local Bayesian and denoising operator can greatly improve the visual effect of the reconstructed image at the cost of a small amount of image details.