Single-shot phase retrieval using automatic differentiation and bilateral total variation

The non-interferometric phase retrieval technique is a versatile tool with a wide range of applications including quantitative phase imaging, astronomical imaging, and beam shaping. The single-shot phase retrieval technique, requiring a single measurement, simplifies optical setups but presents implementation challenges. We propose a single-shot phase retrieval method based on automatic differentiation, along with physical constraints and bilateral total variation regularization. The non-convex optimization problem of phase retrieval can be efficiently solved using a gradient descent algorithm, and the model incorporating prior knowledge effectively characterizes the phase features. Simulation results demonstrate that our method achieves high-quality phase reconstruction with noise robustness. The proposed framework is flexible, easy to extend, and has great potential for addressing various phase retrieval challenges.