手指静脉特征识别技术因其具有高精度、高安全性、活体识别和内部特征的优势受到越来越多的关注。然而，基于二维图像的手指静脉识别面临一个严峻的挑战—静脉特征二义性，即立体分布不同的手指静脉可能存在相同或相似的二维平面投影，降低了手指静脉识别系统的识别精度。本项目从提高手指静脉识别系统鲁棒性出发，结合图像处理和机器学习的知识，研究三维手指静脉识别的深入理论与实现方法，以期建立鲁棒三维手指静脉识别完整理论框架。具体内容包括基于监督编码的手指静脉纹路分割，基于单视图和多视图的手指静脉深度估计，基于相关熵损失的三维静脉特征提取，基于跨域知识迁移的三维静脉特征提取，基于特征层和决策层融合的三维手指静脉特征识别，建立三维手指静脉识别完整理论框架。项目的成功实施，将有望对手指静脉识别系统的性能改善及实际的推广应用产生积极影响，具有重要的理论意义和工程应用价值。

Due to the advantages of high accuracy, high security, liveness recognition as well as inner feature, finger-vein recognition has drawn increasing attention. However, finger-vein recognition faces a serious challenge. The vein features with different 3D distribution structure may be similar in 2D projection plane, which degrades the identification accuracy of finger-vein recognition system. To improve robustness of existing finger-vein recognition system, based on the theory and method in image processing and machine learning, we investigate the theory and implementation method of 3D finger-vein recognition to establish a complete theoretical framework of robust 3D finger-vein recognition. It consists of supervised encoding for finger-vein feature extraction, single and multi-view images based 3D vein depth estimation, the 3D finger-vein feature extraction based on correntropy criterion, the Cross-Domain knowledge transfer for 3D vein feature extraction, the feature-level fusion based 3D finger-vein recognition, the decision-level fusion based 3D finger-vein recognition, as well as the presentation of a theoretical 3D finger-vein system framework. The successful implementation of project will hopefully improve the performance, promotion and application of finger vein recognition. Therefore, this research is important both from theoretical and practical perspectives.