Personalized Multimodal Federated Learning for Fingerprint and Finger Vein Recognition

Hengyu Mu, Jian Guo, Xingli Liu, Chong Han, Lejun Gong

Abstract: Recently, research has demonstrated that multi-biometric recognition techniques are superior in accuracy and security to single biometric recognition. Consequently, multi-biometrics has garnered significant scholarly attention. However, the practical acquisition of biometric data is challenging owing to privacy concerns and cost constraints, which imposes significant constraints on research and application in the biometrics field. Recent explorations into the feasibility of distributed biometrics have met with considerable success. These studies have somewhat mitigated the data dependency of biometric recognition systems by implementing distributed biometric recognition methods. However, the majority focus on single biometric modalities. The absence of multimodal feature extraction and fusion techniques prevents the application of these studies to multi-biometric recognition. This paper proposes a personalized multimodal federated learning framework to address multi-biometric recognition challenges. The framework facilitates distributed fingerprint and finger vein recognition, with personalization settings enhancing the global model's adaptability to local data distributions. Additionally, this paper introduces an innovative Multiscale Attentional Feature Fusion (MS-AFF) module for multimodal feature fusion. Extensive experiments are conducted on the publicly available NUPT-FVP dataset. The results demonstrate that our method surpasses existing methods in terms of accuracy and robustness.

External IDs:dblp:conf/icic/MuGLHG24