Residual orientation modeling for fingerprint enhancement and singular point detection
Introduction
As an important feature in fingerprint images, ridge orientation pattern plays a critical role in fingerprint image enhancement [1], [2], [3], singularity characterization [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], fingerprint classification [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], fingerprint indexing [24], [25], [26], [27], [28], fingerprint recognition [29], [30], [31], [32], etc. There have been a large number of research efforts towards the reliable estimation of fingerprint orientation pattern from acquired fingerprint images, which can roughly be classified into two categories: local estimation and global modeling.
For local estimation methods, the orientation at a pixel is derived based on the information in a neighborhood of the pixel. The most frequently used local method is gradient estimation, which firstly calculates the gradient using the gradient operator (such as the Sobel operator) in digital image processing. Then the orientation is simply the direction perpendicular to the gradient. Despite of its numerical efficiency, the gradient operator is known to be sensitive to noise. To address this issue, a low-pass filter can be applied to the estimated orientation field for noise removal. Alternatively, one can resort to more sophisticated methods, for example, filter-bank [33], [34], statistical techniques [35], structure tensor [36], [37], [38], local voting [39], integration operator [40] and ridge projection [41].
In practice, the quality of acquired fingerprint can easily be degraded for reasons like wet finger, dry finger and finger with presence of crease, wound or scar. Under these circumstances, the structure of fingerprint in a local region can be very weak and the local signal to noise ratio can be low, leading to difficulty for reliably estimating the ridge orientation by local estimation methods. In general, the fingerprint orientation field is sufficiently smooth except for a few points with singularities, thus it is possible to infer local structure using more global information. Pioneered in this direction is the zero-pole model by Sherlock and Monro [42], where singular points, cores and deltas, are modeled as zeros and poles in the complex plane, and the orientation is estimated by the summation of the influence of singular points. This model has received a number of interests and there have been several improvements. Vizcaya and Gerhardt [43] improved this zero-pole model to deal with more degree of freedom around the singular points. Gu et al. [44], [45], [46] propose a combination model for orientation field representation, in which the global orientation is firstly constructed by a polynomial model and subsequently corrected by a point-charge model in regions near singular points. A similar idea has been presented in [47]. Very recently, a unified model is presented in [48] where the zero-pole model and its various generalizations can be regarded as special cases.
In spite of impressive results presented in the above works, these global modeling methods have a common limitation, i.e., they all require the prior knowledge on singular points in the acquired fingerprints. However, fingerprint singular point detection by itself is a nontrivial issue in the characterization of fingerprints, which depends very much on the quality of the fingerprint image. For good quality fingerprints, Poincare index method would suffice in the localization of singular points. But for poor quality fingerprints, there would be a large number of spurious singular points if a simple singular point detection method, such as the Poincare index method, is in use. For most of aforementioned global modeling methods, singular points are often detected manually, which evidently limits their application to realistic system. In [49], an SVM classifier is employed to remove the spurious points detected by the Poincare index method, thus avoiding the manual detection for each fingerprint image. However, a set of training data with manually labeled singular points is necessary before the method can be used. In view of this problem, Wang et al. [50] present a fingerprint orientation model which fits the orientation field using a set of trigonometric polynomials. The method does not require the prior knowledge on singular points and has also been demonstrated to be advantageous over the combination model in fingerprint image enhancement and fingerprint matching. The method has recently been extended in [51], where Legendre polynomial is utilized and a step of singularity preservation using the Levenberg–Marquardt algorithm to minimize the modeling cost functional is introduced after the initial modeling. This method is advantageous in preserving singular points, but at cost of computation load. In addition, the step of minimizing the modeling cost helps to preserve true singularities, but at times some false singularity could be kept as well.
In this paper, we will propose a method for fingerprint orientation reconstruction. The method basically consists of two phases: a preliminary modeling phase by a polynomial regression model, followed by a refined phase for fingerprint regions around singularities. What is different from the combination model is that the proposed method does not require the prior knowledge of fingerprint singularities. And more importantly, instead of having a fixed model for region with singularity, the model for region with singularity in the proposed method is updated through an iterative process, where the singular region is gradually determined from the analysis of the residual field between the original orientation field and the global orientation model. The process is fully automatic and robust against various perturbations. Compared with existing polynomial regression model (in particular, the recently published Legendre polynomial model [51]), the proposed method addresses the issue of singularity preservation and differentiates the true singularity from the singularity due to various artifacts in order to reduce the false alarm rate and to preserve true singularities.
The rest of the paper is organized as follows. Section 2 gives a brief account on polynomial modeling for fingerprint orientation reconstruction, which serves as a preliminary modeling in the proposed method. After that, details on orientation residual analysis are given in Section 3, including singularity region detection as well as the refined model. In Section 4, experiment for validating the proposed method and comparison with the state-of-the-art are presented. Finally the paper is concluded in Section 5 with a discussion.
Section snippets
Preliminary orientation modeling
A recent trend in fingerprint orientation modeling is to fit the orientation field using a set of basis functions, such as polynomial basis, and Fourier basis. Usually the calculation is carried out in the cosine and the sine domain other than in the original orientation field directly. In addition, the orientation angle is doubled before the sine/cosine operation to avoid the problem of orientation cancellation [33]. For completeness, a brief account is given in the following. Firstly, let us
Analysis of residual orientation field for orientation refining
For notational convenience, let us denote the transformed orientation field as
Thus, the analysis of the orientation field can be carried out in the complex domain. In this section, we will focus on the analysis of the residual orientation field zresidual, which is the discrepancy between the original orientation field and the reconstructed onewhere |▪| denotes the magnitude of complex number and is the reconstructed orientation field in the complex
Experimental results
To evaluate the performance of the proposed residual orientation modeling method (ROM), the fingerprint databases FVC 2004 Db1 and Db2 [54] are employed in this study, which contain 800 fingerprint images (100 fingers and each with 8 impressions) per database and are acquired using an optical sensor and a thermal sweeping sensor respectively. The data are collected with alternating pressure and distortion, as well as dried and moisten condition.
Since it is difficult to have the “ground truth”
Discussion
This paper has presented a novel method for fingerprint orientation modeling, which executes in the following two phases. Firstly, the orientation field is fitted to a lower order polynomial model in order to capture the global orientation pattern in the fingerprint structure. Next, the preliminary model around the region with presence of fingerprint singularities is adaptively refined using a higher order polynomial model. Among which, the singularity region is automatically detected through
Acknowledgments
This work was supported by Department of Electrical Engineering, Kasetsart University, Thailand Research Fund (TRF) through the Royal Golden Jubilee Ph.D. Program (Grant no.PHD/0017/2549), and Institute for Infocomm Research (I2R), A-star, Singapore. Prof. Josef Bigun is appreciated for discussion on structure tensor and its application to orientation modeling and fingerprint enhancement. The authors also thank Sharat Chikkerur for sharing of his K-plet fingerprint matching code.
Suksan Jirachaweng is currently a Ph.D. student with Kasetsart University, Thailand. His research interest is in image processing, pattern recognition and various applications.
References (55)
- et al.
Fingerprint enhancement using STFT analysis
Pattern Recognition
(2007) - et al.
Fingerprint pattern classification
Pattern Recognition
(1984) - et al.
Localization of corresponding points in fingerprints by complex filtering
Pattern Recognition Letters
(2003) - et al.
Definition and extraction of stable points from fingerprint images
Pattern Recognition
(2007) - et al.
Fingerprint analysis and singular print detection
Pattern Recognition Letters
(2007) - et al.
Combining flat and structured representations for fingerprint classification with recursive neural networks and support vector machines
Pattern Recognition
(2003) - et al.
Continuous vs exclusive classification for fingerprint retrieval
Pattern Recognition Letters
(1997) - et al.
Fingerprint matching using orientationcodes and polylines
Pattern Recognition
(2007) - et al.
Descriptors for image-based fingerprint matchers
Expert Systems with Applications
(2009) - et al.
Analyzing oriented pattern
Computer Vision, Graphics, and Image Processing
(1987)
Fingerprint classification
Pattern Recognition
Computing oriented texture fields
CVGIP: Graphical Models and Image Processing
Enhanced gradient-based algorithm for the estimation of fingerprint orientation fields
Applied Mathematics and Computation
Extracting image orientation feature by using integration operator
Pattern Recognition
Fingerprint orientation field estimation using ridge projection
Pattern Recognition
A model for interpreting fingerprint topology
Pattern Recognition
A nonlinear orientation model for global description of fingerprints
Pattern Recognition
Modeling orientation fields of fingerprints with rational complex functions
Pattern Recognition
Constrained nonlinear models of fingerprint orientations with prediction
Pattern Recognition
Modeling fingerprint ridge orientation using Legendre polynomials
Pattern Recognition
Crease detection from fingerprint images and its applications in elderly people
Pattern Recognition
Fingerprint image enhancement: algorithm and performance evaluation
IEEE Transaction on Pattern Analysis and Machine Intelligence
Fingerprint enhancement based on discrete cosine transform
ICB
Systematic methods for the computation of directional fields and singular points of fingerprints
IEEE Transaction on Pattern Analysis and Machine Intelligence
Singular points analysis in fingerprints based on topological structure and orientation field
ICB
Fingerprint reference point detection based on local axial symmetry
ICPR
A new detection approach for the fingerprint using extended relational graph
ICIP
Cited by (45)
Combining weighted linear project analysis with orientation diffusion for fingerprint orientation field reconstruction
2017, Information SciencesCitation Excerpt :An increasing number of people have focused on the study of the reliable FOF estimation. A wealth of methods have been proposed for reconstructing the FOF, which can be roughly classified as the gradient based methods [3,17,18,21,28,36], orientation filed modeling methods [4,19,24,27,33,34,37,42], the filter-bank based methods [15,20], the neural network method [29], the ridge projection method [16], and the method based on prior knowledge of fingerprint structure [9]. The most popular method for estimating the FOF is the gradient based method.
A spatial domain scar removal strategy for fingerprint image enhancement
2016, Pattern RecognitionWalking to singular points of fingerprints
2016, Pattern RecognitionFinger-vein pattern restoration with Direction-Variance-Boundary Constraint Search
2015, Engineering Applications of Artificial IntelligenceCitation Excerpt :Biometric personal identification techniques, such as fingerprint (Jirachaweng et al., 2011), facial (Wang and Wu, 2010), iris (Roy et al., 2011), and vein pattern recognition (Wang et al., 2008), have attracted increased attention on various platforms with higher security and lower identification error rate requirements.
A survey of fingerprint classification Part I: Taxonomies on feature extraction methods and learning models
2015, Knowledge-Based SystemsDistortion correction for the orthogonally-splitting-imaging pose sensor
2015, Optics and Laser Technology
Suksan Jirachaweng is currently a Ph.D. student with Kasetsart University, Thailand. His research interest is in image processing, pattern recognition and various applications.
Zujun Hou is a Senior Research Fellow with the Institute for Infocomm Research, A-star, Singapore. His research interest is image processing and its applications, in particular the application to medical image processing and biometrics.
Wei-Yun Yau received his B.Eng. (Electrical) from the National University of Singapore (1992), the M.Eng. degree in biomedical image processing (1995) and Ph.D. degree in computer vision (1999) from the Nanyang Technological University. From 1997 to 2002, he was a Research Engineer and then Program Manager at the Centre for Signal Processing, Singapore leading the research and development effort in biometrics signal processing. His team won the top 3 positions in both speed and accuracy in the international Fingerprint Verification Competition 2000 (FVC2000). Wei Yun served as the Program Director of the Biometrics Enabled Mobile Commerce (BEAM) Consortium from 2001 to 2002. Currently, he is seconded to the Institute for Infocomm Research, leading the research in multi-modal biometrics. He also participates in both the national and international biometric standard activities and is currently the Chair of the Biometrics Technical Committee, Singapore, Chair of the Asian Biometric Forum and project editor of the international standards ISO/IEC JTC1 SC37 29794-4 on fingerprint quality score normalization. Wei Yun is also the recipient of the TEC Innovator Award 2002, the Tan Kah Kee Young Inventors’ Award 2003 (Merit) and Standards Council Merit Award 2005. His research interest includes biometrics, video understanding, homeland security, healthcare monitoring and intelligent systems and has published widely, with 2 patents and 70 publications in these areas.
Vutipong Areekul is an Assoc. Prof. with Kasetsart University, Thailand. His research interest is in image processing, pattern recognition and various applications.