计算机应用 ›› 2015, Vol. 35 ›› Issue (10): 2789-2792.DOI: 10.11772/j.issn.1001-9081.2015.10.2789

• 第十五届中国机器学习会议(CCML2015)论文 • 上一篇    下一篇

结合并行融合的序列化多模态生物特征识别系统框架

李海霞1,2, 张擎3   

  1. 1. 山东省高校证据鉴识实验室(山东政法学院), 济南 250014;
    2. 山东政法学院 信息学院, 济南 250014;
    3. 山东大学 计算机科学与技术学院, 济南 250101
  • 收稿日期:2015-06-01 修回日期:2015-07-03 出版日期:2015-10-10 发布日期:2015-10-14
  • 通讯作者: 张擎(1982-),女,山东济南人,工程师,博士研究生,主要研究方向:生物特征识别、机器学习,zq-abby@163.com
  • 作者简介:李海霞(1976-),女,山东成武人,副教授,博士,主要研究方向:机器学习、图像处理。
  • 基金资助:
    国家自然科学基金资助项目(61173069);山东省高校证据鉴识重点实验室(山东政法学院)开放课题资助项目(KFKT(SUPL)-201409)。

Framework of serial multimodal biometrics with parallel fusion

LI Haixia1,2, ZHANG Qing3   

  1. 1. Evidence Forensic Laboratory in Universities of Shandong Province (Shandong University of Political Science and Law), Jinan Shandong 250014, China;
    2. School of Information, Shandong University of Political Science and Law, Jinan Shandong 250014, China;
    3. School of Computer Science and Technology, Shandong University, Jinan Shandong 250101, China
  • Received:2015-06-01 Revised:2015-07-03 Online:2015-10-10 Published:2015-10-14

摘要: 针对多模态生物特征识别系统并行融合模式中使用方便性和使用效率方面的问题,在现有序列化多模态生物特征识别系统的基础上,提出了一种结合并行融合和序列化融合的多生物特征识别系统框架。框架中首先采用步态、人脸与指纹三种生物特征的不同组合方式以加权相加的得分级融合算法进行的识别过程;其次,利用在线的半监督学习技术提高弱特征的识别性能,从而进一步增强系统的使用方便性和识别可靠性。理论分析和实验结果表明,在此框架下,随使用时间的推移,系统能够通过在线学习提高弱分类器的性能,用户的使用方便性和系统的识别精度都得到了进一步提升。

关键词: 多模态生物特征识别, 序列化集成, 并行集成, 使用方便性, 半监督学习

Abstract: In the multimodal biometric system, the parallel fusion mode has more advantages than the serial fusion mode in convenience and efficiency. Based on current works on serial multimodal biometric system, a framework combined with parallel fusion mode and serial fusion mode was proposed. In the framework, the weighted score level fusion algorithm using biological features of gait, face and finger was proposed at first;then semi-supervised learning techniques were used to improve the performance of weak traits in the system, and the simultaneous upgrade of user convenience and recognition accuracy was achieved. Analysis and experimental result indicate that the performance of the weak classifier can be improved by online learning, the convenience and recognition accuracy are successfully promoted in this framework.

Key words: multimodal biometric, serial fusion, parallel fusion, user convenience, semi-supervised learning

中图分类号: