Robust lensless computational imaging via nonlinear weighting approach

In the framework of multi-image phase retrieval techniques, it is commonly assumed that all diffraction intensity images have the same accuracy and thus contribute equally during the sample reconstruction process. However, it is a fact that the noise and systematic errors inconsistently affect diffraction images in the experiment. In this paper, a nonlinear weighting strategy is introduced into the parallel mode phase retrieval algorithm. During the wavefront updating process, the similarity between the measured amplitude at the reference position and the computed amplitude obtained through diffraction calculations for each plane is used to determine the corresponding weighting factors. When a diffraction pattern is more severely damaged, the similarity decreases, and the related weight coefficient is reduced accordingly. Such a weightings strategy effectively reduces the influence of measurement planes with large errors on phase recovery, thus achieving high-quality reconstruction. Numerical simulations demonstrate that the proposed scheme exhibits excellent robustness, and effectively addresses the deterioration problem of the reconstructed image caused by noise and systematic errors. Ultimately, it successfully reconstructs three different samples in experiments with high accuracy, clarity, and resolution.