PalmKeyNet: Palm Template Protection Based on Multi-modal Shared Key

The distinct ridge features of palmvein and palmprint images, among other palm-related images, make them vulnerable to reversible attacks that can reconstruct the original structure, leading to permanent leakage of biometric features. Additionally, existing multi-modal template protection schemes treat the feature data of each modality as independent, failing to fully capture the inter-modality correlation. Therefore, this paper proposes a multi-modal shared biometric key generation network called PalmKeyNet. By designing keys unrelated to the original palm images as biometric templates, the irreversibility of features is achieved. Additionally, by constructing a multi-modal biometric key generation network, we transform the palm images of different modalities into a unified feature-key space, enhancing the inter-modal correlation. Furthermore, LDPC coding is introduced for multi-modal key error correction to reduce noise interference and improve key discriminability. The proposed approach simultaneously enhances the discriminability, correlation, and security of multi-modal features. The trained PalmKeyNet can be deployed in four modes: single-modal matching (palmprint vs. palmprint and palmvein vs. palmvein), multi-modal matching, and cross-matching. Experimental results on four publicly available palm databases consistently demonstrate the superiority of the proposed method over state-of-the-art approaches.