电子学报 ›› 2021, Vol. 49 ›› Issue (8): 1533-1540.DOI: 10.12263/DZXB.20190473

• 学术论文 • 上一篇    下一篇

三维掌纹局部方向特征二进制编码研究

杨冰1,3, 莫文博2, 姚金良1,3   

  1. 1.杭州电子科技大学计算机学院,浙江 杭州 310018
    2.衢州职业技术学院信息工程学院,浙江 衢州 324000
    3.浙江省脑机协同智能重点实验室,浙江 杭州 310018
  • 收稿日期:2019-04-29 修回日期:2021-02-02 出版日期:2021-08-25
    • 作者简介:
    • 杨 冰(通信作者) 女,1985年生,安徽砀山人.博士,杭州电子科技大学计算机学院副教授,硕士生导师.主要从事模式识别、计算机视觉研究. E-mail:yb@hdu.edu.cn
      莫文博 男,1994年生,吉林通化人.硕士,衢州职业技术学院信息工程学院教师,主要从事掌纹识别、图像处理研究.
    • 基金资助:
    • 国家自然科学基金 (U1909202); 浙江省省属高校基本科研业务费专项资金 (GK209907299001-008)

Learning Local Direction Binary Code for 3D Palmprint

YANG Bing1,3, MO Wen-bo2, YAO Jin-liang1,3   

  1. 1.School of Computer Science and Technology, Hangzhou Dianzi University, Hangzhou, Zhejiang 310018, China
    2.Information Engineering School, Quzhou College of Technology, Quhou, Zhejiang 324000, China
    3.Key Laboratory of Brain Machine Collaborative Intelligence of Zhejiang Province, Hangzhou, Zhejiang 310018, China
  • Received:2019-04-29 Revised:2021-02-02 Online:2021-08-25 Published:2021-08-25
    • Supported by:
    • National Natural Science Foundation of China (U1909202); Fundamental Research Funds for Provincial Colleges and Universities of Henan Province (GK209907299001-008)

摘要:

三维掌纹能显著地减少应用过程中潜在的安全隐患,近年来吸引了越来越多的关注.然而,现有的三维掌纹识别方法大多借助人工设计的描述符来进行匹配,这往往需要先验知识.本文提出一种基于学习策略的局部方向特征二进制编码来完成三维掌纹识别.该方法利用形状指数来描述三维掌纹的局部几何特征,并且在形状指数图像上计算Gabor滤波器响应并将响应差值组合起来形成特征向量.提出利用哈希学习模型得到特征映射函数并将响应差值特征向量转换为方向特征二进制编码,并对方向特征二进制编码图采用分块策略形成特征直方图来进行匹配.在香港理工三维掌纹数据库上的实验结果表明,本文方法在识别率上要优于目前流行的其他三维掌纹识别方法,从而验证了本文方法的有效性.

关键词: 三维掌纹识别, 局部方向特征二进制编码, 形状指数, 哈希学习

Abstract:

Three dimensional (3D) palmprint could efficiently reduce potential security risks in practical applications, and thus attracts a growing interest in recent years. However, most existing 3D palmprint recognition methods require hand-craft designed descriptors and strong prior knowledge are needed. In this paper, we propose a local direction binary code learning method for 3D palmprint recognition. We employ shape index representation to demonstrate the geometry characteristics of local regions in 3D palmprint data, and form Gabor filter response difference vector as the feature vectors by applying the Gabor filter on the shape index image. We utilize the hash learning model to learn feature mapping functions that can project these feature vectors into direction binary code, and further construct the block-wise histograms for matching. Experiments on Hong Kong Polytechnic University 3D palmprint database validate that our method outperforms existing state-of-the-art methods in terms of recognition accuracy, showing the effectiveness of our method.

Key words: 3D palmprint recognition, local direction binary code, shape index, hash learning

中图分类号: