Face Video Steganography for Privacy-protection Automatic Depression Assessment

Xinyi Ni, Zijian Wu, Lu Liu, Siyang Song; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2025, pp. 80-86

Video-based automatic depression assessment (ADA) system have been widely explored and deployed for real-world usage. While face videos recorded for ADA contains sensitive and private contents, existing standard depression face video delivery and storage strategies lack effective privacy protection mechanism. To address this, this paper proposes the first depression face video steganography strategy which enables the invertible concealment of sensitive depression face videos, while ensuring user privacy. We decompose the facial secret video into individual frames and hide each secret frame into a cover image of the same size, enabling high-capacity steganography for video data. Meanwhile, we leverage Stable Diffusion to allow users to generate their own cover videos, enhancing controllability. During the hiding process, we propose a novel Invertible Network to ensure the precise recovery of facial attributes and behavioral information. Experimental results demonstrate that our video steganography method largely preserves the accuracy of the assessment of depression from facial videos. Our code will be provided upon acceptance.