Deep hologram converter from low-precision to middle-precision holograms

Area of Science:

• Computer Vision
• Digital Holography
• Deep Learning

Background:

• Low-precision holograms are computationally efficient but lack detail.
• Increasing hologram precision typically requires significant computational resources.
• Optimizing hologram data for both software and hardware approaches is challenging.

Purpose of the Study:

• To introduce a deep hologram converter utilizing deep learning.
• To convert low-precision holograms into middle-precision holograms.
• To investigate the trade-offs between network size, image quality, and inference speed.

Main Methods:

• A deep learning-based hologram converter was proposed.
• Low-precision holograms were generated using reduced bit width.
• Two deep neural networks (DNNs), one small and one large, were evaluated for the conversion task.

Main Results:

• The large DNN achieved superior image quality for the converted holograms.
• The smaller DNN demonstrated faster inference times.
• The deep hologram converter effectively enhanced hologram precision.

Conclusions:

• Deep learning offers a viable method for enhancing hologram precision.
• The choice between small and large DNNs allows for balancing image quality and processing speed.
• The proposed scheme is adaptable to various hologram calculation algorithms beyond point-cloud methods.