Single-pixel compressive optical image hiding based on conditional generative adversarial network.

Area of Science:

• Optics
• Computer Science
• Cryptography

Background:

• Compressive optical image hiding (COIH) enables secure data transmission.
• Traditional COIH methods often require complex system parameters for reconstruction.
• Single-pixel detectors offer a simplified approach to optical data acquisition.

Purpose of the Study:

• To develop a deep learning framework for COIH using a single-pixel detector.
• To enable direct reconstruction of secret images from compressed signals.
• To improve image reconstruction quality and security in COIH systems.

Main Methods:

• A conditional generative adversarial network (CGAN) was adapted into an end-to-end compressive sensing generative adversarial network (eCSGAN).
• The eCSGAN directly reconstructs 2D secret images from 1D compressive sampling signals.
• The framework bypasses the need for security keys like illumination patterns or host images.

Main Results:

• The eCSGAN achieved high-quality image reconstruction at lower sampling rates compared to traditional compressive sensing (CS).
• The method demonstrated superior image reconstruction quality.
• Numerical simulations and optical experiments validated the feasibility and security of the eCSGAN framework.

Conclusions:

• The developed eCSGAN provides an effective and secure method for compressive optical image hiding.
• This deep learning approach simplifies the COIH process by eliminating the need for system-specific security keys.
• The framework offers a promising advancement in secure optical data transmission and image reconstruction.