Work

Privacy-Region-Aware Adaptive Steganography for Secure Edge-Based Visual Streams

Year: 2026
Type: article
Abstract: With the deployment of Internet of Things (IoT) devices and intelligent surveillance systems in smart cities, edge processing of continuous visual streams is important for reducing privacy leakage during transmission. Existing solutions still struggle to balance security, reversibility, and computation on edge devices. Encryption exposes the existence of protected data, object removal causes permanent information loss, and global steganographic schemes introduce structural rigidity and redundant computation. To address these challenges, this paper proposes a region-aware adaptive reversible steganographic framework for edge-based visual streams. First, we combine scene-consistent inpainting with invertible neural network (INN) embedding, so the generated cover matches the current scene while the original frame remains recoverable. Second, a privacy-region-aware adaptive depth gating mechanism matches the network depth to the scale of privacy objects: deeper layers are used for large regions, and shallow layers are used for small regions to reduce latency. Finally, an intermediate feature alignment strategy transfers invertible mapping rules from a teacher model to lightweight steganographic students. The evaluation further examines threshold sensitivity, temporal fluctuation, and payload-range leakage, showing stable behavior under different operating conditions. Experimental results show that the proposed framework improves edge inference speed on devices such as Jetson Orin Nano Super (67 INT8 TOPS), while maintaining a Peak Signal-to-Noise Ratio (PSNR) above 40 dB for high-fidelity recovery and retaining resistance to steganalysis. These results support scene-consistent, recoverable, and lightweight secure visual-stream transmission on resource-constrained edge devices.
Language: English
Open Access status: hybrid
Estimated APC paid: $2,645
OpenAlex ID: W7204279013
FWCI:
Cites: 26
Cited by:
Related to: 10