Elsevier

Pattern Recognition

Volume 63, March 2017, Pages 87-101
Pattern Recognition

Dense registration of fingerprints

https://doi.org/10.1016/j.patcog.2016.09.012Get rights and content

Highlights

  • A novel dense fingerprint registration algorithm.

  • Dual resolution block based fingerprint registration.

  • Experimental results on three databases containing many distorted fingerprints.

Abstract

Dense registration of different impressions of the same finger is beneficial to various fingerprint matching methods. This is a challenging problem due to elastic distortion of finger skin and sparsity of distinctive features (namely minutiae) in fingerprints. Most existing fingerprint registration algorithms produce only correspondences between minutiae, which are not sufficient for dense registration of fingerprints. In this paper, we proposed a novel dense fingerprint registration algorithm, which consists of a composite initial registration step and a dual-resolution block-based registration step. The dual-resolution block-based registration is approached in an energy minimization framework which consists of local search, energy function construction and global optimization. In local search step, a candidate set of transformations of every input image block are found using image correlation w.r.t. the corresponding reference image block. In energy function construction, two factors are considered: (1) the similarity between the transformed input block and the corresponding reference block, and (2) the compatibility between transformations of neighboring input blocks. In global optimization, a region growing style algorithm is proposed to minimize the energy function. Experimental results on three databases containing many distorted fingerprints, namely FVC2004 DB1, Tsinghua Distorted Fingerprint database and NIST SD27 latent fingerprint database, show that the proposed algorithm not only produces more accurate registration results but also improves the matching performance by fusion of minutiae matching and image correlation.

Introduction

Although automatic fingerprint recognition technology has been widely used in various applications, there is still large room for improvement of matching accuracy, especially in the case of low quality fingerprints [1]. Degradation of fingerprint quality can be photometric or geometrical. Photometric degradation can be caused by non-ideal skin conditions and background noise. Geometrical degradation is mainly caused by skin distortion. Photometric degradation has been widely studied and a number of enhancement algorithms [2], [3], [4], [5], [6] have been proposed. On the contrary, geometrical degradation due to skin distortion has not yet received sufficient attention, despite of the importance of this problem.

Elastic distortion is introduced due to the inherent flexibility of fingertips, contact-based fingerprint acquisition procedure, and a purposely lateral force or torque, etc [7], [8]. Skin distortion increases the intra-class variations, namely the difference among fingerprints from the same finger. Although distortion affects all fingerprint matching methods, image-based matchers are much more sensitive to distortion than minutiae-based matchers. This is one of the reasons for the popularity of minutiae-based matchers. However, the performance of minutiae matchers may drastically drop when the distortion is severe, the number of genuine minutiae is small, or many spurious minutiae are present.

In order to overcome the limitation of minutiae-based matching, researchers have proposed to use some extended features, such as ridge orientation field [9], ridge period map [10], [11], and ridge skeleton [12], [13] etc. Although these methods improve the discriminating ability of minutiae matchers, they also suffer from distortion.

In order to remove the negative impact of distortion, it is necessary to estimate a dense deformation field between two fingerprints. “Dense” means that it can align not only minutiae but also ridges. This is a deformable image registration problem, which is a popular topic in medical image analysis [14], but is first proposed in the fingerprint recognition community. The proposed dense registration algorithm combines the advantages of minutiae-based matching and image-based matching, and overcomes their disadvantages. It can densely register two fingerprints. It is based on a composite initial registration step and a dual-resolution block-based registration step. In initial registration step, the minutiae-based matching can decrease the distortion in global level using the TPS model, which overcomes the disadvantage of image-based matching which cannot handle distortion in global level. In dual-resolution block-based registration step, the idea is to find dense point correspondences between two initial registered fingerprints using local image-based matching. Since the points are regularly sampled grid points, it is not restricted by the sparsity of minutiae. Combining the advantage of minutiae-based matching which can decrease the distortion in global level and the advantage of local image-based matching which can well handel the distortion in local level, the proposed dense registration algorithm can register distorted fingerprints accurately. Registration results for a pair of distorted fingerprints are given in Fig. 1 to compare three different methods, namely, the minutiae-based rigid transformation, the minutiae-based TPS model, and the proposed dense registration method.

The flowchart of the proposed registration algorithm is shown in Fig. 2. Initialized by a minutiae-based registration algorithm, a dual-resolution block matching algorithm is proposed to find dense correspondences between an input fingerprint and a reference fingerprint. Block matching is approached in an energy minimization framework which consists of local search, energy function construction and global optimization. In local search step, a candidate set of transformations of every input image block, which lead to good alignment between the input image block and the reference fingerprint, are found. In energy function construction, due to the continuity of deformation field, two factors are considered: (1) the similarity between the transformed input block and the corresponding reference block, and (2) the compatibility between transformations of neighboring input blocks. In global optimization, a region growing style algorithm is proposed to minimize the energy function. Benefitting from the way to construct energy function and its optimization algorithm, the dual-resolution block-based registration is insensitive to inaccurate initialization and robust to noise.

The proposed algorithm has been evaluated on three databases which contain many distorted fingerprints, namely FVC2004 DB1 [15], Tsinghua Distorted Fingerprint (TDF) database and NIST SD27 latent fingerprint database [16]. Not only the registration accuracy but also matching accuracy are evaluated on these three databases. In order to measure registration accuracy, we have manually registered 120 pairs of distorted fingerprints in TDF as ground truth using a specially designed registration software tool. To quantitatively evaluate the contribution of the proposed dense registration algorithm to matching performance, we conduct matching experiments on fingerprints without/with dense registration. All experimental results demonstrate that the proposed algorithm not only produces more accurate registration results but also improves the matching performance significantly by fusion of minutiae matching and image correlation.

The rest of the paper is organized as follows. In Section 2, we review the related work. In 3 Initial registration, 4 Dual-resolution block-based registration, we present the proposed registration algorithm in details. In Section 5, we give the experiment results. In Section 6, we summarize the paper and discuss the future research directions.

Section snippets

Related work

For a fingerprint registration algorithm, a transformation model is used to register input fingerprint to reference fingerprint. The most common transformation model is the rigid transformation model which contains translation and/or rotation [17], [9]. However, it is obvious that the rigid transformation model cannot handle distorted fingerprints. Thus, some researchers proposed to use more powerful transformation models, such as thin-plate spline (TPS) model [18], [19], to register two

Initial registration

In the initial registration step, a minutiae matching algorithm is used to find matched minutiae between the input and reference fingerprints. If the number of matched minutiae exceeds a threshold tn, a TPS model [22] fitted to the matched minutiae is used to initially register the input fingerprint to the reference fingerprint. Otherwise, more robust features (orientation map and period map) are used to estimate the rigid transformation between two fingerprints. The later method is mainly

Dual-resolution block-based registration

The proposed dual-resolution block-based registration consists of a TPS warping after a low resolution block matching step and a TPS warping after a high resolution block matching step. It is used to further register the input fingerprint to reference fingerprint accurately. The flowchart of dual-resolution block-based registration is shown in Fig. 4.

We define sampling grid on initially registered input fingerprint, and then find the optimal transformation of every sampling block using image

Experiment

The proposed algorithm has been evaluated in terms of registration accuracy (Section 5.1) as well as matching accuracy (Section 5.2) using three databases (see Table 2). The computational cost is discussed in Section 5.3.

Conclusion

Elastic deformation of fingerprints poses a big challenge for dense registration of fingerprints, which is beneficial to various fingerprint matching methods. Most existing fingerprint matching algorithms can output only matched minutiae and hence cannot register severely distorted fingerprints.

In this paper, we proposed a novel dense fingerprint registration algorithm, which consists of an initial registration step and a dual-resolution block-based registration step. The dual-resolution

Acknowledgments

This work is supported by the National Natural Science Foundation of China under Grants 61622207, 61373074, 61225008, 61572271, and 61527808, the National Basic Research Program of China under Grant 2014CB349304.

Xuanbin Si received B.S. degree and Ph.D. degree from the Department of Automation, Tsinghua University, Beijing, China, in 2010 and 2015, respectively. He is now a post doctoral researcher in Graduate School at Shenzhen, Tsinghua University. His research interests include fingerprint indexing, pattern recognition and computer vision.

References (30)

  • A. Senior et al.

    Improved fingerprint matching by distortion removal

    IEICE Trans. Inf. Syst.

    (2001)
  • X. Si et al.

    Detection and rectification of distorted fingerprints

    IEEE Trans. Pattern Anal. Mach. Intell.

    (2015)
  • N. Yager, A. Amin, Evaluation of fingerprint orientation field registration algorithms, in: Proceedings of...
  • D. Wan et al.

    Fingerprint recognition using model-based density map

    IEEE Trans. Image Process.

    (2006)
  • A. Ross et al.

    Fingerprint warping using ridge curve correspondences

    IEEE Trans. Pattern Anal. Mach. Intell.

    (2006)
  • Cited by (48)

    • A Novel Minutiae Triangulation Technique for Non-invertible Fingerprint Template Generation

      2021, Expert Systems with Applications
      Citation Excerpt :

      The EER of the proposed system is compared with some of the latest results reported in literature which follow the FVC protocol and some results from the FVC2004 (FVC2004, 2004). For DB1 the performance is only next to Si et al. (2017). For DB2 and DB3 the performance is better than all the reported results.

    • Pre-registration of translated/distorted fingerprints based on correlation and the orientation field

      2020, Information Sciences
      Citation Excerpt :

      In this step, matching between the registered fingerprint f” and the reference fingerprint r is performed by an image correlator, which outputs the image correlation coefficient, as a fingerprint matcher. Considering that Si et al. [17] introduced the VeriFinger 6.2 SDK [35] in their method for fingerprint matching, we also conducted matching experiments with VeriFinger as one fingerprint matcher. The whole registration algorithm can be iterated, based on the above Eqs. (5), (6), (9) and (23).

    View all citing articles on Scopus

    Xuanbin Si received B.S. degree and Ph.D. degree from the Department of Automation, Tsinghua University, Beijing, China, in 2010 and 2015, respectively. He is now a post doctoral researcher in Graduate School at Shenzhen, Tsinghua University. His research interests include fingerprint indexing, pattern recognition and computer vision.

    Jianjiang Feng is an associate professor in the Department of Automation at Tsinghua University, Beijing. He received the B.S. and Ph.D. degrees from the School of Telecommunication Engineering, Beijing University of Posts and Telecommunications, China, in 2000 and 2007, respectively. From 2008 to 2009, he was a post doctoral researcher in the PRIP lab at Michigan State University. He is an Associate Editor of Image and Vision Computing. His research interests include fingerprint recognition and computer vision.

    Bo Yuan is mostly interested in Data Mining, Evolutionary Computation and Parallel Computing. He received the B.E. degree from Nanjing University of Science and Technology, China, in 1998, and the M.Sc. and Ph.D. degrees from The University of Queensland, Australia, in 2002 and 2006, respectively. From 2006 to 2007, he was a Research Officer on a project funded by the Australian Research Council at The University of Queensland. He is currently an Associate Professor in the Division of Informatics, Graduate School at Shenzhen, Tsinghua University.

    Jie Zhou was born in Nov 1968. He received B.S. degree and M.S. degree both from Department of Mathematics, Nankai University, Tianjin, China, in 1990 and 1992, respectively. He received Ph.D. degree from Institute of Pattern Recognition and Artificial Intelligence, Huazhong University of Science and Technology (HUST), Wuhan, China, in 1995. From then to 1997, he served as a postdoctoral fellow in Department of Automation, Tsinghua University, Beijing, China. From 2003, he has been a full professor in Department of Automation, Tsinghua University. His research area includes computer vision, pattern recognition and image processing. In recent years, he has authored more than 100 papers in peer-reviewed journals and conferences. Among them, more than 30 papers have been published in top journals and conferences such as PAMI, T-IP and CVPR. He is an associate editor for International Journal of Robotics and Automation, Acta Automatica and two other journals. Dr. Zhou is a recipient of the National Outstanding Youth Foundation of China.

    View full text