A provably privacy-preserving image-in-image hiding framework for IoT covert communication

The rapid development of Internet of Things (IoT) has led to the widespread transmission of heterogeneous image data with varying privacy sensitivities over cloud platforms. Highly sensitive images often contain identity-related, biometric, and behavioral information, making them vulnerable to privacy leakage during public dissemination. Image-in-image hiding provides a solution for covert communication by embedding secret images into visually natural cover images. However, existing methods primarily focus on protecting secret data, while largely neglecting privacy leakage risks from the cover images themselves, as stego images must preserve high visual similarity to the covers, exposing sensitive information to identity inference and re-identification attacks. To address this issue, this paper proposes a provably privacy-preserving image-in-image hiding framework for IoT covert communication. Specifically, a DCT-domain block differential privacy mechanism (DCT-BDP) is developed for unstructured data, which achieves an effective trade-off between cover privacy protection and data utility preservation while preserving thumbnails. In addition, a cascaded high-capacity hiding scheme is developed to support multiple secret image embedding with high-quality reconstruction. Theoretical analysis proves that the proposed framework satisfies rigorous (ϵ, δ )-differential privacy guarantees. Extensive experiments demonstrate that the proposed framework achieves high embedding capacity, strong cover privacy protection, and robust resistance to steganalysis attacks while maintaining satisfactory visual quality, thereby providing enhanced security for IoT covert communication.