Latent Vector Optimization-Based Generative Image Steganography for Consumer Electronic Applications

doi:10.1109/TCE.2024.3354824

Latent Vector Optimization-Based Generative Image Steganography for Consumer Electronic Applications

2024-02-01

3

PRE

AI

DOI：10.1109/TCE.2024.3354824

Original

Original request for help

Save

Abstract

En 中文

In consumer electronic applications, to transmit secret images securely, it is required to explore the advanced covert communication technology, i.e., Generative Image Steganography (GIS). However, the existing GIS schemes suffer from the issues of poor stego-image quality and limited hiding capacity. Consequently, these GIS schemes cannot meet the requirements of consumer electronic applications, in which massive secret information needs to be transmitted securely. To address the above issues, we propose a Latent Vector Optimization (LVO)-based GIS scheme, in which the information hiding is implemented by the flow-based generative model during the image generation. Specifically, the LVO algorithm is introduced to compute the hiding probability of each element of latent vector according to its impact on the quality of the stego-image generated from the latent vector. Then, it hides more information in elements with higher hiding probability. The extensive experiments demonstrate that, compared to current GIS schemes, the proposed LVO-based GIS scheme generates higher-quality images, while maintaining hiding capacity (up to 5.0 bpp) and accurate information extraction (almost 100% accuracy rate).

Keywords:

Generative model

generative steganography

AI-generated content

consumer electronics

Latent Vector Optimization-Based Generative Image Steganography for Consumer Electronic Applications

Abstract

Journal

Organization