Realistic blur layer-based computer-generated holography

Abstract

Computer Generated Holography (CGH) based on deep learning has rapidly advanced, surpassing traditional physics-based methods that rely on optical wave simulations and signal processing. We propose a multilayer hologram generation model for generating 3D Phase-only holograms (POHs) using a layer-based approach. The 3D object is represented as multiple layers, with Gaussian convolution kernels applied to generate target images that simulate realistic blur effects for non-focal layers. The model utilizes learnable initial phases to train and optimize the blur effects across these layers. By taking amplitude and depth images as input, the proposed method is capable of synthesizing both 2D and 3D holograms with realistic blur effects. Both simulations and optical experiments demonstrate that the proposed method achieves exceptional hologram generation performance, with blur effects closely matching those observed in real-world scenarios.