Removal of unwanted terms from single shot in-line digital holograms by convolutional neural network

A model to achieve high-resolution three-dimensional microscopic images from synthetically generated digital holograms by using Convolutional Neural Networks (CNNs) is proposed. By employing low-cost microscopy systems and computational techniques, we demonstrate that proposed model provides viable alternative to costly high-resolution microscopic systems. Specifically, the study focuses on the elimination of the unwanted terms in backward-propagated holograms to closely approximate original high-resolution objects.