Real-time 4K computer-generated hologram based on encoding conventional neural network with learned layered phase

Abstract

Learning-based computer-generated hologram (CGH) demonstrates great potential for real-time high-quality holographic displays. However, real-time 4K CGH generation for 3D scenes remains a challenge due to the computational burden. Here, a variant conventional neural network (CNN) is presented for CGH encoding with learned layered initial phases for layered CGH generation. Specifically, the CNN predicts the CGH based on the input complex amplitude on the CGH plane, and the learned initial phases act as a universal phase for any target images at the target depth layer. These phases are generated during the training process of the coding CNN to further optimize the quality. The CNN is trained to learn encoding 3D CGH by randomly selecting the depth layer in the training process, and contains only 938 parameters. The generation time for a 2D 4K CGH is 18 ms, and is increased by 12 ms for each layer in a layered 3D scene. The average Peak Signal to Noise Ratio (PSNR) of each layer is above 30dB in the depth range from 160 to 210 mm. Experiments verify that our method can achieve real-time layered 4K CGH generation.

Generating real-scene hologram through light field imaging and deep learning

“…Compared with non-deep learning methods, they have huge advantages in image generation speed, occlusion detection and image quality. Horisaki et al [9] proposed a non-iterative CGH method based on deep learning; Peng [13] introduced a CITL holographic technology that can achieve real-time 2D holographic display; Shi et al [14] proposed a residual network to quickly generate high-quality 3D holograms; Choi et al [15] proposed a method called Michelson holography that can optimize the image quality of holographic near-eye displays; Zhong et al [16] trained blood deficiency coding 3D CGH by randomly selecting depth layers during the training network, achieving real-time 4 K CGH; Xu et al [17] achieved image reconstruction with only one hologram in digital holography by introducing deep learning.…”