Neural compression for hologram images and videos

Area of Science:

• Optics and Photonics
• Computer Vision
• Immersive Technologies

Background:

• Holographic near-eye displays offer high-quality 3D visuals with focus cues.
• Large data requirements for wide field-of-view and eyeboxes challenge current virtual and augmented reality (VR/AR) applications.
• Existing compression methods are insufficient for complex-valued holographic data.

Purpose of the Study:

• To introduce an efficient deep-learning-based compression method for complex-valued hologram images and videos.
• To address the significant data storage and streaming challenges in VR/AR holographic displays.
• To demonstrate the superiority of the proposed method over conventional codecs.

Main Methods:

• Development of a novel deep learning architecture tailored for complex-valued holographic data.
• Implementation of compression algorithms for both static holographic images and dynamic video sequences.
• Comparative analysis against standard image and video compression techniques.

Main Results:

• The deep-learning method achieves superior compression performance for holographic content.
• Significant reduction in data size for holographic images and videos was demonstrated.
• The method effectively handles the complex-valued nature of holographic data.

Conclusions:

• Deep learning offers a powerful solution for compressing holographic data.
• This advancement can enable more practical and widespread adoption of holographic VR/AR systems.
• The proposed compression technique overcomes key limitations of current holographic display technologies.