High resolution partial fingerprint alignment using pore-valley descriptors
High resolution partial fingerprint alignment using pore-valley descriptors
复制标题
DOI：
10.1016/j.patcog.2009.08.004
复制
发表时间：
2010-03-01
影响因子：
8
通讯作者：
Luo, Nan
中科院分区：
文献类型：
--
作者：
Zhao, Qijun;Zhang, David;Luo, Nan
本文讨论了高分辨率部分指纹的对准问题，这是部分指纹识别的关键步骤。现有的指纹比对方法，包括基于细节特征和基于非细节特征的指纹比对方法，由于较小的指纹片段往往不具备这些方法所需的足够特征，因此不适合部分指纹比对。在本文中，我们提出了一种基于孔隙的高分辨率部分指纹对齐方法，孔隙是一种指纹细脊特征，即使在小的指纹区域也很丰富。首先采用高斯差分滤波方法从指纹图像中提取孔隙。在孔隙检测的基础上，提出了一种新的孔隙谷描述符（PVD），根据孔隙的位置和取向以及孔隙周围的脊状取向场和谷状结构对孔隙进行表征。在此基础上，提出了一种基于pvd的粗孔到细孔匹配算法来定位孔隙对应。一旦确定了相应的孔隙，就可以估计出两个部分指纹之间的对准变换。利用所建立的高分辨率部分指纹数据集和两个指纹匹配器，将该方法与基于代表性特征和基于方向场的方法进行了比较。实验结果表明，基于pvd的方法可以更准确地定位相应的特征点，估计对齐变换，从而显著提高高分辨率部分指纹识别的精度。2009爱思唯尔有限公司版权所有。
This paper discusses the alignment of high resolution partial fingerprints, which is a crucial step in partial fingerprint recognition. The previously developed fingerprint alignment methods, including minutia-based and non-minutia feature based ones, are unsuitable for partial fingerprints because small fingerprint fragments often do not have enough features required by these methods. In this paper, we propose a new approach to aligning high resolution partial fingerprints based on pores, a type of fingerprint fine ridge features that are abundant on even small fingerprint areas. Pores are first extracted from the fingerprint images by using a difference of Gaussian filtering approach. After pore detection, a novel pore-valley descriptor (PVD) is proposed to characterize pores based on their locations and orientations, as well as the ridge orientation fields and valley structures around them. A PVD-based coarse-to-fine pore matching algorithm is then developed to locate pore correspondences. Once the corresponding pores are determined, the alignment transformation between two partial fingerprints can be estimated. The proposed method is compared with representative minutia based and orientation field based methods using the established high resolution partial fingerprint dataset and two fingerprint matchers. The experimental results show that the PVD-based method can more accurately locate corresponding feature points, estimate the alignment transformations, and hence significantly improve the accuracy of high resolution partial fingerprint recognition. (C) 2009 Elsevier Ltd. All rights reserved.