Study On Phase Retrieval Algorithm Based On Sparse Representation And Cartoon-texture Model

Phase retrieval is an issue that tries to recover an image from its Fourier magnitude measurements. Since the Fourier magnitude measurements contain less information, the traditional phase retrieval algorithms can not reconstruct the image efficiently under the scenario that the oversampling ratio is relatively low. Therefore, how to use the suitable image priors to improve the reconstruction quality of the image is the key issue. This paper focuses on phase retrieval algorithms, the specific contents are as follows:Firstly, the paper proposes a novel reconstruction algorithm based on the sparsity of image over the discrete cosine transform. Compared with the traditional algorithms on the ability of reconstruction and denoising performance. The experiments show that the quality of image reconstruction has been improved.Secondly, proposing the the phase retrieval algorithm based on the total variation, for which can represent image contour information and edge information effectively. Proposing the the phase retrieval algorithm based on the dual-tree complex wavelet, for which can represent image texture information well. In order to make the algorithm has robustness, error term is introduced in that algorithms. Compared with the traditional algorithms, The quality of image reconstruction has further improved and robust for noise.Finally, considering that ingle sparse priori can not reconstruct the image efficiently under the scenario that the oversampling ratio is relatively low. In this paper, the cartoon-texture model is utilized for phase retrieval algorithm. Two sparse representation methods including both total variation and dual-tree complex wavelet are exploited to decompose the image into two parts, namely the cartoon component and the texture component, proposing the phase retrieval algorithm based on cartoon-texture model. The experimental results show that the proposed algorithm can effectively improve the quality of image reconstruction and robust for noise.