Related Experiment Video
Updated: May 6, 2026
10:16
Digital Inline Holographic Microscopy DIHM of Weakly-scattering Subjects
Published on: February 8, 2014
12.7K
Enhancing light efficiency in phase-only holograms via neural network.
Balakiruthika Periyasamy1, Heeseong Hwang1, Daeho Yang2
1Department of Physics, Gachon University, 1342 Seongnam-daero, Sujeong-gu, Seongnam-si, Gyeonggi-do, Korea.
Scientific Reports
|October 14, 2025
Summary
ShuffleResnet, a novel neural network, enhances hologram reconstruction efficiency by 59% and improves image quality. This artificial neural network approach offers faster processing for real-time holographic applications.
Area of Science:
- Optics and Photonics
- Computer Vision
- Artificial Intelligence
Background:
- Conventional double phase encoding methods (DPM) for phase-only spatial light modulators (SLMs) face limitations in light efficiency and reconstruction quality.
- Artificial neural networks (ANNs) show promise in advancing hologram synthesis and reconstruction.
Purpose of the Study:
- To introduce ShuffleResnet, a neural phase encoding approach to overcome DPM limitations.
- To enhance light efficiency and reconstruction fidelity in holographic applications.
Main Methods:
- Developed ShuffleResnet, a neural phase encoding model.
- Utilized numerical simulations to evaluate performance against conventional DPM.
- Tested hologram encoding at 1920x1080 resolution.
Main Results:
- Achieved a 59% increase in light efficiency compared to DPM.
- Demonstrated improved reconstruction quality and artifact suppression.
- Attained an average inference speed of 4.74 milliseconds per hologram.
Conclusions:
- ShuffleResnet significantly enhances light efficiency and reconstruction fidelity.
- The model's fast inference speed indicates strong potential for real-time holographic applications.
- This neural network approach offers a superior alternative for phase-only SLM-based holography.
