Palmprint recognition system for mobile device based on circle loss

Abstract

The extensive use of mobile devices such as smartphones has made the research of identification on mobile devices a prominent topic. Palmprint recognition is one of the promising but underestimated technologies suitable for mobile devices. In this paper, we propose a novel palmprint recognition method towards mobile devices by Incorporating MobileFaceNet and Circle loss (IMC). MobileFaceNet can speed up operation while maintaining high accuracy using a lightweight model. Circle loss has a more flexible optimization strategy and a clearer convergence boundary than other loss functions like softmax, which can further improve the accuracy. Adequate experiments conducted on several mobile palmprint datasets in the XJTU-UP database show the superiority of our method. The Equal Error Rate (EER) on IF dataset is decreased to 2.88%. In comparison to other palmprint recognition methods, our proposed IMC method can obtain more promising performance, where the EERs are decreased to 3.94% on HN dataset and 3.29% on HF dataset.