High-fidelity lensless imaging with single hologram based on physics-guided frequency-aware network

Lensless in-line holography is widely used to obtain depth information on pathological tissues and biological cells to enable noninvasive analysis, due to low cost and large field-of-view. However, existing systems suffer from crosstalk dilemma of twin-image that limit their capability to achieve high-resolution phase imaging with a single measurement, especially in the presence of noise. In this Letter, we propose a physics-guided frequency-aware network (FA-PGNet) for single-shot lens-free holographic reconstruction. The FA-PGNet method uses frequency-level supervision to balance the high- and low-frequency information between images and gradually obtained reliability results by incorporating a physical model. With the co-addition of frequency-aware and physics-guided, it can easily tackle the twin-image and speckle noise, with 5.88 dB improvement on the peak signal-to-noise ratio and 6-fold reduction on the uncertainty of the imaging system. Optical experiments demonstrate that the FA-PGNet allows access to complex features of microscopic cells with high resolution. We believe that the proposed method provides a powerful tool for single-shot cell imaging.