Phase retrieval refers to reconstructing the unknown original image or signal which is by using the amplitude-only measurement data of its linear transform.The phase retrieval in the optical field refers to the phase distribution of the reconstructed light wave.The process of reconstructing image or the optical wave phase is a linear transformation inverse problem,since the amplitude measurement data does not include the phase information part,the phase retrieval is a typical ill-posed problem.During the actual phase retrieval process,the measurement data is always disturbed by a variety of noises,so it is a hot issue to study the phase retrieval algorithm which has robustness to noise.In this paper,based on the existing phase retrieval algorithms,we add plug-and-play priori knowledge to the phase retrieval process,proposing phase retrieval algorithm based on plug-and-play prior,the main researches as follows:Firstly,the algorithm of error reduction shows that the algorithm has the advantages of fast convergence and low computational complexity.However,when the image oversampling rate is low,the algorithm can not reconstruct the image completely.Considering that the measurement data is polluted by noise,while image denoising method can be used to improve the image quality,so in the phase retrieval algorithm,the amplitude knowledge,constraint support,dual-domain filter and so on can be used as priori information at the same time.In the near-field diffraction model,this paper proposed a phase retrieval algorithm,named phase retrieval algorithm based on dual-domain image denoising prior.The experimental result shows that the algorithm reconstruct image with high quality,and has the advantages of robustness and fast convergence.Secondly,for the practical application,the phase retrieval process is always polluted by noise,a phase retrieval algorithm based on BM3D image filtering is proposed for reconstructing image.The algorithm uses the priori information including amplitude constraint,the non-negative support information of image space domain and the BM3D filtering prior,then introduces the l2 norm model to enhance the robustness of noise.The experimental result shows that the algorithm can reconstruct the image with high resolution under different intensity noises when the oversampling rate is lower than 3.Finally,based on the maximum a posteriori estimation theory and the modular structure feature of the alternating direction multiplier method,any existing image denoising method can be added into phase retrieval algorithm,the alternating direction multiplier method is used to solve the phase retrieval problem.In this paper,we proposed a phase retrieval algorithm based on plug-and-play prior.Among this algorithm,the image denoising method is BM3D,and the directional multiplier multiplication method is used to solve the optimization problem.The prior information can be divided into three kinds:the transform domain amplitude prior,the non-negative constraint and the plug-and-play prior.The experimental result shows that the algorithm achieves effective realistic reconstructed images visually under low sampling rate and different intensity noises. |