Deep D2C-Net: deep learning-based display-to-camera communications.

Area of Science:

• Computer Science
• Electrical Engineering
• Image Processing

Background:

• Display-to-camera (D2C) communication enables data transfer via visual signals.
• Existing D2C methods face challenges with optical wireless channel robustness and data embedding quality.

Purpose of the Study:

• To introduce Deep D2C-Net, a novel D2C communication technique leveraging deep convolutional neural networks (DCNNs).
• To develop end-to-end encoding and decoding networks for high-quality data embedding and robust data acquisition.
• To evaluate the performance of Deep D2C-Net against state-of-the-art algorithms.

Main Methods:

• Developed Deep D2C-Net with fully end-to-end encoding and decoding networks.
• Introduced Hybrid layers for encoding, concatenating feature maps of data and cover images.
• Utilized a simple convolutional neural network (CNN) for decoding.
• Conducted real-world experiments using smartphone cameras and digital displays under varying parameters (distance, angle, brightness, resolution).

Main Results:

• Deep D2C-Net demonstrated superior performance compared to existing algorithms.
• Achieved higher peak signal-to-noise ratio (PSNR) and lower bit error rate (BER).
• Generated data-embedded images with high visual quality for human observers.

Conclusions:

• Deep D2C-Net offers a robust and high-quality solution for display-to-camera communication.
• The proposed DCNN-based approach significantly improves data embedding and extraction efficiency.
• The technique shows promise for various optical wireless communication applications.