Robust Skin Color Driven Privacy Preserving Face Recognition via Function Secret Sharing
This work addresses privacy concerns in face recognition systems for applications like security and authentication, though it appears incremental as it builds on existing privacy-preserving methods with specific enhancements.
The paper tackles privacy-preserving face recognition by using skin color patches to train auxiliary feature extractors and recognition models, achieving improved performance while being robust against black-box attacks and GAN-based restoration. They propose a Function Secret Sharing-based embedding comparison protocol that prevents intermediate result leakage and shows greater efficiency than Secret Sharing-based protocols.
In this work, we leverage the pure skin color patch from the face image as the additional information to train an auxiliary skin color feature extractor and face recognition model in parallel to improve performance of state-of-the-art (SOTA) privacy-preserving face recognition (PPFR) systems. Our solution is robust against black-box attacking and well-established generative adversarial network (GAN) based image restoration. We analyze the potential risk in previous work, where the proposed cosine similarity computation might directly leak the protected precomputed embedding stored on the server side. We propose a Function Secret Sharing (FSS) based face embedding comparison protocol without any intermediate result leakage. In addition, we show in experiments that the proposed protocol is more efficient compared to the Secret Sharing (SS) based protocol.