Tomographic Reconstruction Based On Phase Retrieval

Tomographic reconstruction based on phase retrieval has become a typical method for 3D phase object visualization.And the accuracy of the tomography of phase objects is affected by many factors.One of the most important influencing factors is undoubtedly the accuracy of the phase projection and the position of rotation center of the Tomographic reconstruction.On the one hand,the phase of the object is obtained by the phase retrieval algorithm,and the classical phase retrieval algorithms including the iterative method and the transport of intensity equation have some accuracy problems,such as the uncertainty of the iteration,the approximation of the intensity derivative,the mechanical error and lens noise in the intensity acquisition,etc.The application range of different phase retrieval algorithms will also be limited by the distance factor,which will directly affect the final 3D reconstruction accuracy;on the other hand,laminar reconstruction requires projection in all directions,which is usually obtained by rotating the object,while in practice the object is often not accurately placed in the center of rotation.Therefore,it will have a great impact on the reconstruction accuracy.In response to the above mentioned issues,the paper has carried out corresponding research.The main research work and innovations are as follows:(1)A convolutional neural network-based phase retrieval algorithm at all defocus distance is proposed.The algorithm mainly includes two important components: phase retrieval and convolutional neural network optimization,among which,the phase retrieval includes three modules: intensity transfer equation-based phase retrieval algorithm,angular spectrum iteration algorithm and hybrid iteration.Different processing modules can be selected according to different out-of-focus distances,and high-precision phase results can be obtained at different out-of-focus distances.Then convolutional neural networks are used to further improve the accuracy of phase retrieval results,and at the same time,the stability and robustness of the phase retrieval algorithm are improved.(2)A Tomographic reconstruction algorithm based on the center of rotation correction is proposed.Firstly,the phase is solved from a set of intensity projection images in each direction by the phase retrieval algorithm,then the phase projection in each direction is converted into a sinusoidal image to calculate the true center of rotation,and finally the calculated center of rotation position is substituted for the original image center,and the object 3D volume is reconstructed slice-by-slice by using the filtered inverse projection algorithm.