Homomorphic Encryption-Based Privacy Protection for Palmprint Recognition

Palmprint recognition is a promising biometric technology. Currently, the research on palmprint recognition focuses on the topic of feature extraction and matching. However, due to the characteristic that biometric traits cannot be modified at will, how to secure and protect the privacy of palmprint recognition is a neglected and valuable topic. In this paper, we propose a privacy-preserving framework for palmprint recognition based on homomorphic encryption. Specially, for any given eligible palmprint recognition network, it is encrypted layer by layer to obtain both image key and network key. Particularly, the introduced homomorphic encryption strategy does not cause any loss of recognition accuracy, which ensures the wide applicability of the proposed method. In addition, it greatly reduces the risk of exposing plaintext images and model parameters, thus circumventing potential attacks against data and models. Adequate experiments on constrained and unconstrained palmprint databases verify the effectiveness of our method.