Scalable Federated Learning for Fingerprint Recognition Algorithm

Fingerprint recognition technology is widely used in various terminal devices and serves as a powerful and effective method for authentication. Existing research relies on centralized training models based on datasets, overlooking the privacy and heterogeneity of the data itself, resulting in the leakage of user information and decreased recognition accuracy. In order to solve the problem of data security and privacy protection, this paper proposes a federated learning-based architecture called FedFR(Federated Learning-Fingerprint Recognition). The parameters from each endpoint are iteratively aggregated through federated learning to improve the performance of the global model under privacy constraints. Moreover, to solve the client-side unfairness issue in traditional federated learning caused by randomly selecting aggregation weights, a client selection method based on reservoir sampling is proposed, increasing the diversity of data distribution. Using the real-world databses, the effectiveness of FedFR is compared and analyzed through simulation experiments. The results show that FedFR exhibits good performance in terms of privacy protection levels, evaluation accuracy, and scalability. Distinct from traditional fingerprint recognition algorithms, FedFR improves the security and scalability of the model from the data source, providing a reference for the application of federated learning in biometric technology.