Computer-generated holography using hybrid planar-spherical wave primitives

Holographic near-eye displays offer immersive 3D viewing with minimal discomfort; however, encoding holographic data remains challenging due to the trade-off between wave modelling accuracy and computational complexity. We introduce a hybrid method combining planar and spherical wave models to synthesize holograms from light fields and depth, accurately modeling both diffuse and specular reflections. Numerical simulations show improved recreation of deep scenes and defocus blur. To address computational complexity, we employ a machine learning framework for creating functional light field data sets from random 3D scenes with proper utilization of the available spatial and angular bandwidths. Our detailed ablation study results demonstrate that the proposed method can be efficiently accelerated via machine learning, thus enabling real-time synthesis of immersive, high-quality holograms.