Fast high-precision hologram generation based on low-precision hologram enhanced by deep learning

Fast computer-generated holography (CGH) calculation is a critical issue in holographic three-dimensional (3D) display. Analytical methods can be used to generate high-precision holograms but the amount of calculation is very heavy. Deep-learning-based CGH methods have made significant progress but they are difficult to deal with 3D point-cloud object model. We propose a novel scheme that combines a simplified analytical method, accurate phase-added stereograms (APAS), with Complex-Valued Convolutional Neural Network (CCNN) to achieve fast generation of holograms from 3D point-clouds. By using APAS to fast generate low-precision holograms from point-clouds, and then using CCNN to restore the degraded holograms, the APAS algorithm can improve the computational speed of hologram generation, and CCNN can ensure the high precision of holograms. By combining the advantages of these two methods, our proposed scheme can achieve a balance between CGH calculation speed and precision. Our proposed scheme is verified by simulated and experimental results.