基于细节点柱形编码和深度卷积特征的指纹检索方法  

Fingerprint Indexing Based on Minutia Cylinder-Code and Deep Convolutional Feature

作　　者：宋德华 封举富[1] 

机构地区：[1]北京大学信息科学技术学院智能科学系机器感知与智能教育部重点实验室,北京100871

出　　处：《模式识别与人工智能》2018年第2期175-181,共7页Pattern Recognition and Artificial Intelligence

基　　金：国家自然科学基金重点项目(No.61333015)资助~~

摘　　要：指纹检索方法使用细节点柱形编码作为特征,充分考虑指纹细节点的局部结构特征,却忽略指纹的整体结构特征,限制指纹检索的准确率.基于此种问题,文中提出基于细节点柱形编码和深度卷积特征的指纹检索方法.使用深度卷积网络学习指纹的整体结构特征(深度卷积特征),并结合深度卷积特征和细节点柱形编码,提升指纹检索的准确率.在3个经典指纹检索数据库上通过实验分析深度卷积特征的特性.实验表明,文中方法有效提升指纹检索的准确率.In the typical fingerprint indexing method based on minutia cylinder-code （MCC） feature, the minutiae local structure is adequately taken into account. Since the global structure of fingerprint is ignored, the accuracy of fingerprint retrieval is limited. Therefore, deep convolutional neural network is employed to learn the global feature （deep convolutional feature） of fingerprint. Then, the MCC and deep convolutional feature are fused to improve the fingerprint indexing accuracy. Experiments are carried out to compare the proposed method with other prominent approaches on three benchmark databases. Besides, the property of deep convolutional feature is analyzed. Experimental results show that the proposed method effectively improves the accuracy of fingerprint indexing.

关 键 词：指纹检索 深度卷积神经网络 细节点柱形编码 特征表示 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]