Phase aberration compensation via a self-supervised sparse constraint network in digital holographic microscopy

Phase aberration compensation is of great significance for the quantitative phase imaging in digital holographic microscopy. Least-square-based fitting algorithms are the most common options except for the double exposure method. However, they either require perturbation hypothesis, image segmentation, or an extra object-free hologram, leading to additional efforts for aberration elimination. In addition, supervised learning-based networks are mostly applied to create object-free masks automatically, but suffer from laborious label preparation and limited generalization capability. In this study, we propose a self-supervised sparse constraint network (SSCNet) for aberration compensation without the perturbation hypothesis, as well as an additional label or mask. Zernike model enhancement and sparse constraints are introduced for fitting aberration with only a single measured hologram. Simulation and experiment demonstrate that our proposed SSCNet can accurately compensate the phase aberration without object-free masks, and is more stable than other algorithms. It can be generalized to other samples with high accuracy and little time requirement. Moreover, SSCNet has advantage in the adaptive compensation of dynamic aberration in continuous measurement. Finally, it is an effective and practical fitting algorithm for phase aberration compensation.