Palm-vein verification based on U-Net

Abstract

As one of the biometric characteristics, palm-vein have received more and more attention in recent years. However, in practical applications, the vein image capturing is affected by various factors, so many low quality images are produced in recognition system. Generally, the vein networks extracted from such low quality images contains many noises, which degrades the recognition accuracy. To address these problems, this paper proposes an end-to-end convolutional neural network to extract vein feature for verification. Firstly, we label the palm-vein pixel by the combination of some handcraft-based palm-vein image segmentation methods and build a training dataset. Secondly, a U-Net network is trained based on the resulting dataset and its outputs are the probability of pixels to belong to vein pattern. Thirdly, we propose a scheme to encode the outputs of U-Net to obtain the vein network patterns. The experiment results on the public CASIA palm-vein dataset implies the effectiveness of our proposed method.