Multitask Identity-Aware Image Steganography via Minimax Optimization.

Area of Science:

• Computer Science
• Information Security
• Digital Image Processing

Background:

• Traditional image steganography risks privacy leakage post-restoration.
• Need for secure methods that protect sensitive data like faces and fingerprints during and after transmission.

Purpose of the Study:

• To propose a novel framework, Multitask Identity-Aware Image Steganography (MIAIS), for high-capacity image steganography.
• To enable direct recognition on container images without restoring secret images, thus enhancing privacy.

Main Methods:

• Developed a framework preserving identity information of secret images within container images.
• Introduced a content loss function for identity preservation and minimax optimization for balancing image similarity and identity protection.
• Incorporated an optional restoration network for a multitask framework.

Main Results:

• Demonstrated robustness transferability across different cover datasets.
• Experimental results show MIAIS outperforms other visual information hiding and high-capacity steganography methods in multitask scenarios.
• Achieved direct recognition on container images, mitigating privacy risks.

Conclusions:

• MIAIS offers a secure and effective solution for high-capacity image steganography with direct recognition capabilities.
• The multitask framework provides flexibility for secret image restoration when needed.
• The proposed method enhances data security by preventing post-restoration privacy leakage.