Automated digital holographic image reconstruction with deep convolutional neural networks

In off-axis digital holographic microscopy, a camera records the spatial interference intensity pattern between light scattered from the specimen and the unperturbed reference light. Digital propagation using the numerical reconstruction algorithm allows both phase-contrast and amplitude-contrast images of the sample to be retrieved. This is possible when the exact distance between the image sensor (such as CCD) plane and image plane is provided. In this paper, we give an overview of our work on a deep-learning convolutional neural network with a regression layer as the top layer to estimate the best focus distance. The experimental results obtained using microsphere beads and red blood cells show that the proposed method can accurately estimate the propagation distance from a filtered hologram. This method can significantly accelerate the numerical reconstruction time since the correct focus is provided by the CNN model with no need for digital propagation at different distances.