We demonstrate how applying adaptive segmentation on context modelling schemes can be used for lossless compression of binary holograms. A tree-based scheme using Bayesian inference is introduced to improve the compression rate over fixed sized models.