Hologram Reconstruction using cascaded deep learning networks

초록

Deep learning technology is one of the emerging topics in solving problems in all scientific fields. In this paper, we address a hologram reconstruction method using cascaded multitask networks. A cascaded network consists of two U-net networks. The first is used for conversion between hologram plane and image plane and the other is used for extraction of image and depth. To train the network, we simulate an optical holographic microscopy setup. Experimental results show that the proposed approach can restore effectively complex optical fields and depth information. © OSA 2021.