Steganographic secret sharing via AI-generated photorealistic images

Abstract

Abstract Steganographic secret sharing is an access control technique that transforms a secret message into multiple shares in a steganographic sense. Each share is in a human-readable format in order to dispel suspicion from a malicious party during transmission and storage. Such a human-readable format can also serve to facilitate data management. The secret can be reconstructed only when a sufficient number of authorized shareholders collaborate. In this study, we use neural networks to encode secret shares into photorealistic image shares. This approach is conceptually related to coverless image steganography in which the data are transformed directly into an image rather than concealed into a cover image. We further implement an authentication mechanism to verify the integrity of the image shares presented in the decoding phase. All coverless image steganography schemes can be used to achieve steganographic secret sharing, but our detection mechanism can further improve the robustness of these schemes. Experimental results confirm the robustness of the proposed scheme against various steganalysis and tampering attacks.

Keywords

• Coverless steganography
• Generative adversarial networks
• Image synthesis
• Secret sharing