Work

DSCCNet for high-quality 4K computer-generated holograms

Year: 2025

Type: article

Abstract: With the increasing demand for high-quality 3D holographic reconstruction, visual clarity and accuracy remain significant challenges in various imaging applications. Current methods struggle for higher image resolution and to resolve such issues as detail loss and checkerboard artifacts. To address these challenges, we propose the model Depthwise Separable Complex-valued Convolutional Network (DSCCNet) for phase-only computer-generated holography (CGH). This deep learning framework integrates complex-valued convolutions with depthwise separable convolutions to enhance reconstruction precision and improve model training efficiency. Additionally, the diffuser is employed to reduce checkerboard artifacts in defocused parts of 3D CGH. Experimental results demonstrate that DSCCNet can obtain 4K images reconstructed with more intricate details. The reconstruction quality of both 2D and 3D layered objects is enhanced. Validation on 100 images from the DIV2K dataset shows an average PSNR above 37 dB and an average SSIM above 0.95. The proposed model provides an effective solution for high-quality CGH applications.

Source: Optics Express

DOI: 10.1364/oe.553575

Authors: Zhenqi Xu, Junmin Leng, Ping Dai, Chao Wang

Institution: Beijing Information Science & Technology University

Language: English

Topic: Advanced Optical Imaging Technologies

Subfield: Media Technology

Field: Engineering

Domain: Physical Sciences

Open Access status: gold

Estimated APC paid: $2,270

Funder: National Natural Science Foundation of China