Robust censored Poisson ptychography with joint framewise gain estimation for severe measurement mismatch

Ptychographic measurements in practice often deviate from the ideal forward model because of detector saturation, framewise intensity drift, and residual background contamination. When these effects appear together, conventional fluctuation correction can become unstable and may lead to biased reconstructions. In this work, we develop a robust ptychographic reconstruction method for such severely mismatched data. Saturated pixels are treated as censored Poisson observations, and framewise gains are estimated jointly within a likelihood-based framework. After background correction, the measurements are incorporated through a pseudo-Poisson projection, while the gains are refined by damped Newton updates with prediction-guided masking and gauge normalization. This design preserves useful information in saturated regions and stabilizes the scale coupling between gains and object amplitude. Simulations and experiments show that, compared with the conventional fluctuation method, the proposed algorithm yields more stable reconstruction under strong clipping and intensity variation, indicating that both saturation-aware modeling and robust gain estimation are essential under severe mismatch.