A one-bit quantization approach for low-dose Poisson phase retrieval

Imaging quality for biological tissue is commonly affected by damages of the specimen caused by illumination particles. To mitigate this issue, often very low doses of illumination have to be used in the experiment. Consequently, the resulting inverse problem is subject to highly noisy data. In this note, we address this issue for the case of diffraction imaging by studying the problem of phase retrieval with low-count Poisson data. Our key idea is to exploit the close connection between the Poisson measurement model and the one-bit quantization problem. We propose a reconstruction method based on algorithmic approaches to that problem and compare the performance of this method with state-of-the-art algorithms for noisy phase retrieval, observing superior performance in a number of relevant examples.