Conversion of 2D picture to color 3D holography using end-to-end convolutional neural network
Chenliang Chang, Chenzhou Zhao, Bo Dai, Qi Wang, Jun Xia, Songlin Zhuang, Dawei Zhang
Abstract
Abstract In the field of holographic 3D display, generating a three-dimensional (3D) computer-generated hologram (CGH) from a single two-dimensional (2D) image has been a significant challenge due to the high-dimensionality of the problem. In this paper, we introduce an end-to-end Convolutional Neural Network (CNN) framework, trained using a large dataset, which directly infers a full-color 3D CGH from a single 2D picture. The proposed method bypasses the need for depth or any other 3D information, facilitating the transformation of readily available 2D images into 3D holograms. We demonstrate that our end-to-end CNN can successfully convert either computer graphics (CG) generated 2D image or real-world captured 2D image into high-quality phase-only hologram, and experimentally achieving the effect of full-color 3D holographic display. Our work extends the horizons of lower-dimensional to higher-dimensional holographic wavefront information conversion, and therefore has potentials to advanced applications such as 3D display technology and metaverse development.