A cross-database attack in biometric systems is a security attack where attackers attempt to leverage a compromised target user's template in one database and infer their templates in other databases. The biometric community largely ignores cross-database attack although they pose poses potential severe risks to the security of the biometric systems. This paper presents a comprehensive study on cross-database attacks in palmprint recognition systems. We specifically focus on the coding-based palmprint templates due to their popularity. Coding-based methods for palmprint feature representation are designed differently to improve performance accuracy and reduce complexity, where the coded templates look completely diverse; thus, it is difficult to correlate them in a meaningful way. However, we demonstrate that the latent correlation of coding-based palmprint templates can indeed be established. Specifically, we analyze six coding-based palmprint representations, and by exploiting the latent statistical correlations among them, we devise an effective cross-database attack algorithm. The attack enables the target users’ palmprint templates to be exploited to infer their templates stored in other databases despite different coding methods. Our cross-database attack even yields a 100% success rate in some scenarios on the public datasets. This suggests high risks of cross-database attacks and privacy invasion of palmprint recognition systems that adopt coding-based representation.
|Publication status||Published - 2023 Mar 15|
Bibliographical noteFunding Information:
This work is funded by the National Natural Science Foundation of China under Grant 61866028 , Technology Innovation Guidance Program Project of Jiangxi Province (Special Project of Technology Cooperation), China under Grant 20212BDH81003 , National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP) under Grant NRF-2022R1A2C1010710 , University of Macau ( MYRG2018-00053-FST ), Sichuan Science and Technology Program, China under Grant 2021JDJQ0024 , and Sichuan University ‘From 0 to 1’ Innovative Research Program, China under grant 2022SCUH0016
© 2023 Elsevier B.V.
All Science Journal Classification (ASJC) codes
- Management Information Systems
- Information Systems and Management
- Artificial Intelligence