RAMN: Reversible animation mask network for face privacy protection

In today’s digital world, the sharing of images containing personal facial information has become a prevalent behavior. But it is essential to recognize that facial information, being a prominent individual characteristic, has found applications in various security domains. Consequently, the inadvertent exposure of facial information can potentially result in significant losses for users. Existing facial privacy solutions aim to modify facial features to avoid recognition by facial recognition systems. However, in some special application scenarios, authorized users (such as family members) hope to restore the original facial information. Therefore, To achieve face privacy protection and the ability to recover original facial information, this paper proposes a face privacy protection method based on the reversible animation mask network. We start by generating the high-resolution anime mask from the real face using an Anime Converter. Then, the real face is embedded within the anime mask as secret information and generates the anime face, which closely resembles the appearance of the anime mask. Authorized users can then extract the original face information from the anime face, and obtain the restored face which looks remarkably similar to the original face. Experiments show that this method has good visibility and can effectively protect the user’s real face privacy. In addition, this method can almost perfectly recover real face information from the anime face.