Comprehensive deep learning model for 3D color holography.

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Holography is crucial for microscopy, imaging, and data encryption, but current methods for image generation and reconstruction are slow.
• Accurate, fast, and versatile techniques are needed for computing holograms and reconstructing object information, especially for color imaging across multiple planes.

Purpose of the Study:

• To design optical holograms for generating multi-plane, multi-color holographic images using a deep learning model.
• To develop a deep learning model capable of reconstructing object information from intensity-based holographic images without phase or amplitude data.

Main Methods:

• A deep learning model, CHoloNet, was designed to generate optical holograms.
• CHoloNet was trained to produce holographic images at multiple observation planes and colors by tuning holographic structures.
• The model was developed to reconstruct object/hologram information directly from intensity images.

Main Results:

• CHoloNet successfully generated optical holograms capable of multiplexing color holographic image planes.
• The deep learning model accurately retrieved object/hologram information from intensity images, showing excellent agreement with ground-truth data.
• Reconstruction did not require iterative processes or multiple holographic images from various observation planes.

Conclusions:

• CHoloNet offers a fast, efficient, and accurate framework for both generating and reconstructing holographic images.
• The developed model accelerates the implementation of optical holography in fields such as microscopy, data encryption, and communication.
• This work contributes a versatile deep learning approach to advance optical holography applications.