Palmprint Biometrics: Online Learning with Differential Evolution and Contrastive Representation

Contactless palmprint biometrics offer a promising solution for mobile authentication with distinctive features of palmprints, such as principal lines and wrinkles. However, a significant challenge arises with the incremental class problem, as users can register their palmprints on their mobile devices at any time without centralized control. This dynamic enrollment creates difficulties in integrating new classes without degrading the system’s performance on existing classes. To address this, our paper present an online evolutive learning approach that combines contrastive learning with a modified differential evolution algorithm, allowing the system to efficiently incorporate new biometric data without necessitating complete model retraining. Utilizing Siamese networks, we develop robust embedding representations that facilitate accurate user registration and authentication. Evaluations on the 11 k Hands dataset demonstrate that our approach significantly outperforms traditional fine-tuning methods, achieving higher accuracy, precision, recall, and F1 score as the number of classes increases. These results highlight the efficacy and practicality of our solution for real-time biometric systems, providing enhanced security and adaptability for mobile applications.