Refined axial localization in holographic particle tracking via Gaussian mixture model-based noise classification

This study resolves trajectory fragmentation in holographic particle tracking, a long-standing problem in axial localization using refocusing criterion. The axial elongation of reconstructed particles makes localization susceptible to artifacts, thus resulting in outliers. Our method models axial errors with a two-component Gaussian Mixture Model to probabilistically identify outliers. It then uses isotropic in-plane statistics to adaptively regularize the trajectory. Evaluations on simulated holograms demonstrate superior robustness to high particle densities over conventional techniques, achieving a mean absolute error of less than the particle diameter in tracking 100 µm particles. The method's practical effectiveness was confirmed through experiments using high-speed holograms of water droplets.