BLANKET: Anonymizing Faces in Infant Video Recordings

2025-12-17

AI Summary

To address the challenge of anonymizing infant faces in videos—requiring simultaneous identity removal, preservation of critical facial attributes, and natural expression dynamics—this paper proposes the first two-stage anonymization framework specifically designed for infant video data. The first stage employs a diffusion model to generate semantically consistent synthetic faces; the second stage performs keypoint-guided temporal-consistent face swapping and expression transfer, reinforced by landmark consistency constraints to ensure inter-frame expression coherence. Evaluated on an infant video dataset, our method significantly outperforms DeepPrivacy2: achieving 100% identity unrecognizability, improving facial attribute preservation by 32%, reducing downstream pose estimation error by 41%, and exhibiting no perceptible artifacts in human evaluation. This work constitutes the first systematic effort to jointly optimize privacy protection, task robustness (e.g., pose estimation), and visual fidelity in infant video anonymization.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Privacy
• Natural Language Processing: Safety and Robustness

Application Category

• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Ensuring the ethical use of video data involving human subjects, particularly infants, requires robust anonymization methods. We propose BLANKET (Baby-face Landmark-preserving ANonymization with Keypoint dEtection consisTency), a novel approach designed to anonymize infant faces in video recordings while preserving essential facial attributes. Our method comprises two stages. First, a new random face, compatible with the original identity, is generated via inpainting using a diffusion model. Second, the new identity is seamlessly incorporated into each video frame through temporally consistent face swapping with authentic expression transfer. The method is evaluated on a dataset of short video recordings of babies and is compared to the popular anonymization method, DeepPrivacy2. Key metrics assessed include the level of de-identification, preservation of facial attributes, impact on human pose estimation (as an example of a downstream task), and presence of artifacts. Both methods alter the identity, and our method outperforms DeepPrivacy2 in all other respects. The code is available as an easy-to-use anonymization demo at https://github.com/ctu-vras/blanket-infant-face-anonym.

Problem

• Anonymizing infant faces in videos while preserving facial attributes
• Ensuring ethical use of infant video data through de-identification
• Maintaining temporal consistency and expression transfer in anonymization

Innovation

• Generates random faces via diffusion model inpainting
• Swaps faces with temporal consistency and expression transfer
• Outperforms DeepPrivacy2 in de-identification and attribute preservation