Phase Retrieval and Hologram Reconstruction Using a Neural Network

The ability to recover the missing phase information of waves has been transformative for a wide set of fields, such as material science and life sciences, leading to numerous scientific discoveries so far. Here, we demonstrate that a deep convolutional neural network can achieve phase retrieval and holographic image reconstruction, using a single intensity-only hologram of a complex sample, which conventionally required multiple measurements for phase recovery. Using a trained convolutional neural network, phase recovery and holographic image reconstruction of a complex-valued sample are performed in a single feed-forward pass, considerably reducing the measurement and the computation time when compared to measurement diversity based phase recovery approaches. We validated this approach by imaging various samples including blood and Pap smears and tissue samples. The results demonstrate the potential of convolutional neural networks for solving challenging inverse problems in computational imaging field.