指静脉识别中若干关键问题研究

Characterized by the distinctive advantages such as live-body acquisition, hard to be forged and stolen, finger vein recognition has a broad market prospect. In recent years, finger vein recognition has become a hot research issue in biometrics. Great progress has been achieved through more than ten years’ endeavors from domestic and foreign colleagues. However, there are still some challenging problems to be solved. Rather than focusing on consensus challenges, this project chooses and concentrates on three important yet not have been noticed problems of finger vein recognition. The three problems are as following: 1. how to take use of the traditional "noisy" information such as deformation and pose variation of the finger vein image to further improve the performance of the finger vein recognition system; 2. how to effectively mine and apply the previously ignored unmatched information to further improve the performance of the finger vein recognition system; 3. how to realize the hierarchical feature description to further promote the comprehensiveness and discrimination ability of feature expression. The three key problems to be investigated are basically general problems in biometric recognition researches, thus, if they are properly solved, corresponding ideas are expected to be applied to other biometric recognition techniques to improve their performance.

因为拥有只能活体采集、难以被伪造和窃取等独特的优势，指静脉识别技术具有广阔的市场前景，近年来已成为生物特征识别的研究热点之一。在国内外同行的十多年的努力下，指静脉识别的研究已经取得很大进展，但仍有较多挑战性的问题需要去解决。本项目并未着眼于指静脉识别中较为共识性的问题，而是选择和凝练了三个较为重要但尚未引起同行足够关注的较新的问题作为切入点，展开相应的研究工作：1、如何利用形变、姿态变化等传统意义上“噪声”信息提高指静脉识别系统性能的问题；2、如何有效挖掘和利用当前普遍被忽略的未匹配信息进一步提升指静脉识别系统性能的问题；3、如何实现指静脉图像的层次化特征描述、进一步提高静脉特征表达的全面性和区分性的问题。本项目所研究的三个关键问题，在生物特征识别方向均具有一定的共性。如能得到有效解决，思路可望用于其他生物特征识别技术，提升其他生物特征识别系统的性能。

按照预定工作计划，项目组重点开展了如下三个方面的工作：.1、研究了传统意义上“噪声”（形变、姿态变化等）信息的反向利用问题，实现了基于形变信息、基于双向形变金字塔和基于几何形状形变等特征提取方法，相关成果发表在SCIENCE CHINA Information Sciences2018、Neurocomputing2018、MM2019、AAAI2019、ICASSP2017等国际期刊和会议。.2、研究了被忽略的未匹配信息的进一步深入挖掘和利用问题，实现了基于区分性二值码学习、基于特征点关系挖掘的特征学习和基于解剖学结构的特征提取等方法，相关成果发表在IEEE TIFS2018、TCSVT2018、PR2017、CVPR2017、ICCV2019等国际期刊和会议。.3、研究了层次化融合特征描述问题，提出了基于张量的异构信息融合、基于超像素上下文融合和基于多方向编码融合等模型，相关成果发表在IEEE TMM2017、TIFS2018、Information Science2019、IJCAI2019、SIGIR2017等国际期刊和会议。.上述工作，有效提高了手指静脉识别系统的性能，并可用于解决其他类似的模式识别问题。.项目基本按预定计划顺利开展，实现了预定研究目标。依托本项目，共发表论文50篇，其中CCF A类 8篇，B类 12篇，C类21篇。申请国家发明专利5项。获得山东省科技进步二等奖1项（项目负责人为第一完成人）培养博士研究生4人，硕士研究生17人。项目研发的部分手指静脉核心算法已在合作企业的相关产品中得到应用。