Cancelable Face Recognition using Deep Steganography

While the dawn of digital privacy fears strikes hard at the very thread of our existence, biometrics, one of the traditional systems, is at risk of invasion through privacy breaches and identity theft. This is because cancellable biometric systems promise through revocation and reissuance of biometric templates. Based on this opportunity, the present work proposes a novel approach in cancellable face recognition through deep steganography such that biometric data is embedded in digital images to protect user privacy while maintaining the highest possible recognition accuracy. The approach utilizes deep learning models to design effective steganographic encodings of facial features that will then be securely embedded into innocuous images. In any given scenario, the embedded features can be extracted and used for a face recognition, thereby not leaking the original biometric data. The steganographic process is reversible, so the original face template can be revoked and replaced with a new one if compromised. We test the proposed system on publicly available face datasets and check the recognition accuracy, steganographic robustness, and cancelability of the proposed method. The results show that the deep steganography-based approach obtains high recognition accuracy to compare with traditional face recognition systems but also provides an extra layer of security by having the cancelability. This is highly potent in improving the privacy and security of biometric systems.