Exploiting holographically encoded variance to transmit labelled images through a multimode optical fiber

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Image Processing

Background:

• Multimode fibers (MMFs) are explored for image transmission using deep learning.
• Deep neural networks (DNNs) decode speckle patterns but underutilize MMF capacity with homogeneous data.
• Existing methods struggle with high-fidelity reconstruction for diverse datasets.

Purpose of the Study:

• To enhance the transmissive capabilities of MMF-DNN systems.
• To encode additional variance into output speckle patterns using holographic modulation.
• To achieve high-fidelity image reconstruction from complex MMF transmissions.

Main Methods:

• Implemented holographic modulation by adding labels to datasets and injecting phase images into the MMF.
• Utilized a ResUNet architecture for decoding holographic speckle patterns.
• Applied holographic labels to segment and reconstruct color images (RGB components).

Main Results:

• Speckle pattern datasets were successfully clustered by holographic labels.
• High-fidelity reconstruction of images was achieved without loss.
• Color images were reconstructed by decoding distinct holographic labels for each RGB component.

Conclusions:

• Holographic modulation significantly improves MMF-DNN system capacity.
• The proposed method enables robust, high-fidelity image transmission and reconstruction.
• This approach offers a novel solution for transmitting complex data, like color images, without temporal synchronization.