摘要
Coherent diffraction microscopy determines the three-dimensional (3D) structure of a noncrystalline sample by acquiring a set of diffraction patterns at multiple sample orientations. In order to achieve a high spatial resolution, it usually requires a large number of diffraction patterns with small increments of tilt angles. However, this is not easily attainable for biological or other radiation-sensitive samples due to the radiation damage issue. With the constraint that only a limited number of diffraction patterns is available, a new method is proposed to optimize the effect of these data for three-dimensional image reconstruction. The method combines the concepts of selecting angles with a constant slope increment, the pseudopolar Fourier transform, and the guided hybrid input-output method for direct 3D phase retrieval. Based on simulated and experimental data, it is shown that the proposed method has a better accuracy than some existing three-dimensional image reconstruction techniques. The method is also examined for its stability and consistency by reducing data points used for image reconstruction. The relationship between the number of diffraction patterns used and the accuracy of the reconstructed images is also studied. © 2011 American Physical Society.