Learned holographic light transport: invited.

Area of Science:

• Optics and Photonics
• Computer Vision
• Machine Learning

Background:

• Computer-generated holography (CGH) algorithms often exhibit discrepancies with physical holographic display outputs.
• Simulations typically do not account for the complex light transport phenomena inherent in real-world holographic systems.

Purpose of the Study:

• To develop a method that accurately models the light transport in holographic displays.
• To improve the fidelity of computer-generated holograms by learning the physical characteristics of display devices.
• To enhance the image quality and simulation accuracy for holographic display applications.

Main Methods:

• A dataset was generated by capturing image reconstructions from optimized holograms displayed on a physical device using a camera.
• A complex-valued convolution kernel was learned, inspired by ideal simulations.
• This kernel effectively propagates generated holograms to match captured photographs.

Main Results:

• The learned kernel significantly improves the accuracy of holographic simulations.
• The method demonstrates a dramatic enhancement in the image quality of holographic displays.
• The approach successfully bridges the simulation-to-reality gap for holographic light transport.

Conclusions:

• Physically informed machine learning approaches can effectively model complex optical phenomena.
• The developed method offers a pathway to more realistic and accurate computer-generated holography.
• This work advances the practical application and performance of holographic display technologies.