An Enhanced EM Framework for Scanning Position Refinement in Ptychography via MCMC Sampling and Posterior Approximation

Ptychography is a powerful imaging technique capable of achieving sub-angstrom resolution in electron microscopy. In biomacromolecule imaging, accurate scanning position refinement becomes essential when significant sample drift occurs, as it is critical to maintaining high image quality. To enhance refinement accuracy, we propose an Expectation Maximization (EM) framework combined with Markov Chain Monte Carlo (MCMC) sampling. By modeling scanning positions as random variables, our method avoids local minima, accelerates convergence, and improves robustness against noisy data. Experiments on synthetic biomacromolecule data, incorporating Poisson noise and poor initial scanning positions, demonstrate that the proposed method outperforms existing algorithms in both accuracy and convergence speed, providing a more reliable solution for biomacromolecule ptychographic imaging where sample drift is prevalent.