Generating Multi-Depth 3D Holograms Using a Fully Convolutional Neural Network.

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Generating 3D holograms efficiently remains a significant challenge in holography.
• Existing methods often struggle with accurate depth representation and occlusion handling.

Purpose of the Study:

• To introduce an efficient method for generating multi-depth phase-only holograms using a fully convolutional neural network (FCN).
• To improve the quality of 3D hologram reconstruction, particularly regarding occlusion relationships and depth focusing.

Main Methods:

• A forward-backward-diffraction framework was employed to compute multi-depth diffraction fields.
• A layer-by-layer replacement method (L²RM) was utilized to manage occlusion relationships.
• A fully convolutional neural network (FCN) was designed to generate multi-depth holograms from diffraction fields.

Main Results:

• The proposed method generates multi-depth 3D holograms with a PSNR of 31.8 dB.
• Hologram generation is achieved rapidly, taking only 90 ms for high-resolution images (2160 × 3840).
• Numerical and experimental results demonstrate accurate reconstruction of clear 3D scenes with correct occlusion and excellent depth focusing.

Conclusions:

• The FCN-based method offers an efficient and effective solution for multi-depth 3D hologram generation.
• The integration of diffraction computation and layer-by-layer replacement enhances reconstruction quality and occlusion handling.
• This approach advances the field of holography by enabling faster and more accurate 3D hologram synthesis.