Abstract
Gabor holography is an amazingly simple and effective approach for three-dimensional (3D) imaging. However, it suffers from a DC term, twin-image entanglement, and defocus noise. The conventional approach for solving this problem is either using an off-axis setup, or compressive holography. The former sacrifices simplicity, and the latter is computationally demanding and time-consuming. To cope with this problem, we propose a model-based holographic network (MB-HoloNet) for three-dimensional particle imaging. The free-space point spread function (PSF), which is essential for hologram reconstruction, is used as a prior in the MB-HoloNet. All parameters are learned in an end-to-end fashion. The physical prior makes the network efficient and stable for both localization and 3D particle size reconstructions.
Highlights
H OLOGRAPHY is a powerful tool for 3D imaging and display due to its wavefront encoding ability
Physics constraints can be explicitly enforced in each stage/iteration, making the whole network inherently physics-aware. Such combined approaches have been successfully applied to computed tomography (CT) [24] and magnetic resonance imaging (MRI) reconstruction [25]
We further show that the network trained with synthesized data works for experimental captured hologram reconstruction, as a natural consequence of the physically motivated property of our proposed learning framework
Summary
H OLOGRAPHY is a powerful tool for 3D imaging and display due to its wavefront encoding ability. Physics constraints can be explicitly enforced in each stage/iteration, making the whole network inherently physics-aware Such combined approaches have been successfully applied to computed tomography (CT) [24] and magnetic resonance imaging (MRI) reconstruction [25]. We integrate the physical principles of holography with deep learning-based approaches for 3D holographic particle imaging. MB-HoloNet takes a single Gabor hologram along with the free-space PSF and outputs the 3D particle volume It is composed of a fixed number of stages, each of which strictly corresponds to an iteration in the shrinkage-thresholding algorithm. We further show that the network trained with synthesized data works for experimental captured hologram reconstruction, as a natural consequence of the physically motivated property of our proposed learning framework. FWe believe MB-HoloNet will serve as an initial point for future unrolled networks for holographic reconstruction
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
