Cancelable Face Recognition with Mask

Cancelable biometrics are the effective methods to solve the problem that biometrics cannot be reissued after leakage. As far as we know, the existing methods of cancelable biometrics are implemented by irreversible transformation or encryption of biometrics. In this paper, we propose a method that we add a mask to face image to realize cancelable biometrics. We use relative entropy to study the statistical relationship between the face images with mask and original face images. Then we change the positions of same mask to achieve the same effective cancelable biometrics while keeping the statistical relationship invariant as far as possible. In addition, we divide the whole mask into blocks, which can be used as cancelable biometrics template as same as the whole one. Similarly, we also maintain the statistical relationship between the original face images and the face images with the blocks as close as possible to the statistical relationship between the face images with the whole one and the original face images. Our experiments on CanBiHaT and LFW dataset show that our method achieve cancelable face recognition.