Deep residual network with data consistency for subsampled Fourier ptychographic microscopy

Fourier Ptychographic Microscopy (FPM) is a computational imaging technique which reconstructs super-resolved amplitude and phase images by combining variably illuminated low-resolution images through an iterative phase retrieval algorithm. However, the phase-retrieval-based reconstruction requires sufficient overlap between spatial frequency bands of the measurements, which creates a trade-off between the number of measurements and the reconstruction quality. We propose a deep-learning-based FPM reconstruction that recovers both amplitude and phase images in high resolution with far fewer measurements than conventional FPM, with model-based constraint. Our model works with almost no overlap between low-resolution measurements in the Fourier domain, only taking into account the total Fourier extent of the measurements.