High-quality real-time 3D holographic display for real-world scenes based on the optimized layered angular spectrum method

Holographic display is ideal for true 3D technology because it provides essential depth cues and motion parallax for the human eye. Real-time computation using deep learning was explored for intensity and depth images, whereas real-time generating holograms from real scenes remains challenging due to the trade-off between the speed and the accuracy of obtaining depth information. Here, we propose a real-time 3D color hologram computation model based on deep learning, realizing stable focusing from monocular image capture to display. The model integrates monocular depth estimation and a transformer architecture to extract depth cues and predict holograms directly from a single image. Additionally, the layer-based angular spectrum method is optimized to strengthen 3D hologram quality and enhance model supervision during training. This end-to-end approach enables stable mapping of real-time monocular camera images onto 3D color holograms at 1024×2048 pixel resolution and 25 FPS. The model achieves the SSIM of 0.951 in numerical simulations and demonstrates artifact-free and realistic holographic 3D displays through optical experiments across various actual scenes. With its high image quality, rapid computational speed, and simple architecture, our method lays a solid foundation for practical applications such as real-time holographic video in real-world scenarios.