I spent spare moments during the break writing code to do diffraction microscopy with few-photon and single-photon images. It works! The estimator has the properties I expect. You could, in principle, do diffraction imaging with many images taken at unknown angles, each of which contains only a single photon! Sounds crazy? Just the asymptotic property of maximum-likelihood estimators. Of course, as the number of photons per image drops, the sensitivity of the result to assumptions (the necessarily wrong assumptions) becomes very strong.