Multi-Level Reversible Facial Privacy Protection Based on Compressive Sensing and Generative Adversarial Network

The cloud platform is an advanced and modernized information delivery, storage and sharing service, and one of its main challenges is to ensure the security, confidentiality, integrity and availability of user data. In order to avoid the potential damage of data leakage, in this paper,we propose a multi-level reversible face privacy preservation system based on compressive sensing and generative adversarial network. The system generates the face of the original image into the face of the anime through the dual style generative adversarial network,embeds the anime face into the original image to generate the secret image. Compressive sensing provides the encryption of the secret image,and the watermark is embedded in the encrypted compressive sensing measurement values, finally generates a cipher image. Fully authorized users can extract the watermark and recover the original image, while semi-authorized users can only recover the secret image. Extensive experimental results indicate the proposed facial privacy preservation method provides a low-cost encryption system with low computational complexity and time complexity. While achieving slightly higher reconstruction quality, the runtime of competing system is nearly twice that of the system presented in this paper.