Few-Shot Learning for Palmprint Recognition via Meta-Siamese Network

Palmprint is one of the discriminant biometric modalities humans. Recently, deep learning-based palmprint recognition algorithms have improved accuracy and robustness results to a new level. Most them require large amount labeled training samples guarantee satisfactory performance. However, getting enough data difficult due time consumption privacy issues. Therefore, in this article, novel meta-Siamese network (MSN) proposed exploit few-shot learning for small-sample recognition. During each episode-based iteration, few images are selected as sample query sets simulate support testing test set. Specifically, model trained episodically with flexible framework learn both feature embedding similarity metric function. In addition, two distance-based losses introduced assist optimization. After training, can ability get scores between testing. Adequate experiments conducted on several constrained unconstrained benchmark databases show that MSN obtain competitive improvements compared baseline methods, where best be up 100%.