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

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

Automated summary

The article introduces a novel meta-Siamese network for small-sample palmprint recognition, which learns feature embedding and deep similarity metric function through episodic training and introduces two distance-based losses for optimization. Experimental results show that the method has competitive advantages in palmprint recognition, with the highest accuracy reaching 100%.