Fig. 12: Learning-based holography.

With the explosive growth of mathematical optimization and computing hardware, deep learning has become a tremendously powerful tool to solve challenging problems in holographic imaging, including every key process of reconstruction. a A conceptual architecture of DNN in DH. b The DNN architecture for holographic reconstruction¹⁶⁵, the input is the reconstructed amplitude and phase from back propagation, and the output is the guess amplitude and phase of the object. c Phase unwrapping using DNN³²⁷, the compared results are performed using modified Goldstein’s algorithm³³⁷. d The CNN-based approach simultaneously performs autofocusing and phase recovery to significantly extend the DOF and the reconstruction speed in holographic imaging³³¹. e The DNN trained for microscopic imaging, and MTF comparison for the input image and the output image of a DNN that is trained on images of a lung tissue section³³². f Viability states of the individual cells from the reconstructed phase of it based on DNN³⁴⁷. g Label-free fluorescence multiplexing of multiple subcellular structures from RI distribution of cell³⁴⁸