Deep compression network for enhancing numerical reconstruction quality of full-complex holograms

Area of Science:

• Optics and Photonics
• Computer Vision
• Signal Processing

Background:

• Digital holography has advanced significantly, but large data sizes hinder commercialization.
• Traditional codecs struggle with the interferometric nature of digital holograms, limiting coding efficiency.
• Existing research often overlooks the quality of numerical reconstruction, which is crucial for viewer experience.

Purpose of the Study:

• To evaluate the coding performance of JPEG2000 and HEVC-Intra for digital holograms.
• To analyze the limitations of current compression standards for holographic data.
• To propose a novel deep learning-based compression network for full-complex digital holograms.

Main Methods:

• Evaluation of JPEG2000 and HEVC-Intra coding performance on digital holograms.
• Development of a deep learning compression network incorporating hologram and numerical reconstruction quality.
• Comparison of the proposed network against JPEG2000, HEVC, VVC, and JPEG-XL.

Main Results:

• JPEG2000 and HEVC-Intra show limitations in compressing digital holograms effectively.
• The proposed deep learning network achieves superior coding efficiency compared to VVC and JPEG-XL.
• The network demonstrates better objective coding efficiency and visual quality than existing methods.

Conclusions:

• A deep learning approach can overcome the limitations of traditional codecs for digital hologram compression.
• Considering numerical reconstruction quality during training is vital for effective hologram compression.
• The proposed network offers a promising solution for efficient digital hologram compression and commercialization.