基于模态信息度评估策略的掌纹掌静脉特征识别方法  

Palmprint and palm vein recognition method based on modal information evaluation strategy

作　　者：潘在宇 徐家梦 王军 贾伟[2] PAN Zaiyu;XU Jiameng;WANG Jun;JIA Wei(School of Information and Control Engineering,China University of Mining and Technology,Xuzhou 221116,China;School of Computer Science and Information Engineering,Hefei University of Technology,Hefei 230009,China)

机构地区：[1]中国矿业大学信息与控制工程学院,江苏徐州221116 [2]合肥工业大学计算机与信息学院,安徽合肥230009

出　　处：《智能系统学报》2024年第5期1136-1148,共13页CAAI Transactions on Intelligent Systems

基　　金：新一代人工智能重大项目(2020AAA01073000);中央高校基本科研业务费专项资金项目(2023QN1077).

摘　　要：多模态生物特征识别技术凭借其出色的识别效果和稳健的可靠性引起了业界的广泛关注。然而,传统多模态生物特征识别方法通常直接在特征层或决策层进行融合,很少考虑模态样本质量不同导致的融合效果差异。此外,由于缺乏大规模公开多模态生物特征数据库,多模态生物特征识别方法的研究受到一定程度上的限制。因此,设计了一款手部多模态数据采集设备,并自建了手部多模态数据库,用于多模态生物特征识别方法的验证与评估;提出了一种基于模态信息度评估策略的掌纹掌静脉特征识别方法,利用样本标签对应的类别置信度来评估每个模态特征的信息度,从而使模型在融合过程中根据不同模态对身份识别的贡献率进行自适应的权重分配。实验表明该方法在2个公开的数据库以及自建数据库上均取得了最高识别率。Multimodal biometric recognition has gained widespread attention in the industry due to its excellent performance and robust reliability.However,traditional multimodal biometric recognition methods usually fuse directly at the feature or matching layer and rarely consider the differences in fusion effects caused by the quality of modal samples.Moreover,research on multimodal biometric recognition methods is restricted by the absence of large-scale publicly available multimodal biometric databases.Therefore,first,a hand multimodal data acquisition device is designed,and a hand multimodal database is created for the validation and evaluation of multimodal biometric recognition methods.Second,a palmprint and palm vein fusion recognition method is proposed based on a modal information evaluation strategy.It uses the category confidence level corresponding to ground truth sample labels to assess the information level of each modal feature.Thus,the model adaptively assigns weights according to the contribution rates of different modes during the identity recognition fusion process.This method outperforms other recognition methods by achieving the highest recognition rate on two public databases and one self-built multimodal biometric database.

关 键 词：生物特征识别 掌纹图像 掌静脉图像 多模态生物特征数据库 模态信息度评估策略 类别置信度 多模态融合 掌纹掌静脉特征识别 

分 类 号：TP30[自动化与计算机技术—计算机系统结构]