The U-Net-based phase-only CGH using the two-dimensional phase grating.

Area of Science:

• Computational optics
• Holography
• Deep learning applications

Background:

• Traditional phase-only hologram generation often struggles with clear diffraction orders.
• U-Net architectures are powerful for image generation but require optimization for specific holographic applications.
• Two-dimensional phase gratings offer unique modulation capabilities for diffraction fields.

Purpose of the Study:

• To generate phase-only holograms with distinct first diffraction orders.
• To optimize U-Net for efficient and high-resolution hologram generation.
• To leverage the properties of 2D phase gratings for improved holographic reconstruction.

Main Methods:

• Investigated the modulation effect of 2D phase gratings on diffraction patterns.
• Adapted U-Net training to target the first diffraction order instead of the zeroth.
• Implemented a 'phase recombination' technique to enhance U-Net efficiency (reduced memory, increased speed).

Main Results:

• Successfully generated 4K resolution phase-only holograms in just 0.05 seconds.
• Achieved an average Peak Signal-to-Noise Ratio (PSNR) of approximately 37.2 dB on the DIV2K-valid-HR dataset.
• Demonstrated the effectiveness of targeting the first diffraction order for clearer holographic images.

Conclusions:

• The proposed method effectively generates high-quality phase-only holograms with clear first diffraction orders.
• The integration of 2D phase gratings and optimized U-Net significantly improves hologram generation speed and quality.
• This approach offers a promising advancement for real-time holographic display technologies.