GCL-MIH: A Generative-Based Coverless Multi-Image Hiding Method.

Area of Science:

• Computer Science
• Information Security
• Digital Image Processing

Background:

• Existing image hiding methods struggle with a trade-off between capacity and security/distortion.
• Cover-based methods offer high capacity but risk image distortion and security vulnerabilities.
• Secure coverless methods provide good security but suffer from limited data embedding capacity.

Purpose of the Study:

• To propose a novel generative-based coverless multi-image hiding method (GCL-MIH) for enhanced security and capacity.
• To enable hiding multiple secret images within a single generated image without requiring a predefined cover image.
• To overcome the limitations of existing methods by achieving high capacity and high security simultaneously.

Main Methods:

• A feature reverse module compresses multiple secret images into feature vectors.
• Vectors are normalized to conform to a standard normal distribution.
• An invertible generative network (Flow-GAN) generates a realistic face image embedding the secret information.

Main Results:

• The GCL-MIH successfully hides up to four secret images within one generated face image.
• Achieved a maximum embedding rate of 32 bits per pixel (bpp), significantly exceeding existing coverless methods.
• Generated stego images exhibit high realism (FID score: 11.98) and recovered images show satisfactory fidelity (avg. PSNR: 33.18 dB, avg. SSIM: 0.9412).

Conclusions:

• The proposed GCL-MIH method effectively addresses the limitations of current image hiding techniques.
• It offers a promising solution for secure and high-capacity secret information transmission using generative models.
• The method demonstrates superior performance in terms of capacity, security, and realism of generated images.