A fingerprint verification algorithm using tessellated invariant moment features
Introduction
Fingerprint-based biometrics systems are often used as the automatic fingerprint verification system (AFVS) for criminal identification and police work. In an AFVS, the input includes a user identity (ID) and a fingerprint, and the output indicates whether the input fingerprint is consistent with the ID. The system simply compares the input fingerprint with the one addressed by the ID in the database. The performance of an AFVS in terms of matching accuracy and computation speed depends mainly on methods of feature extraction and matching algorithm. Researches have been extensively explored on feature extraction and matching algorithms, yet these are still challenging for better performance. Fingerprint feature extraction and matching methods may be broadly classified into three categories: minutiae-based, image-based, and hybrid [16].
The most popular and widely used methods are minutiae-based [10], [13]. These use a feature vector extracted from fingerprints as sets of points in a multi-dimensional plane. The feature vector may comprise several characteristics of minutiae such as type, position, orientation, etc. A typical minutiae-based method essentially searches for the best alignment between the template and the input minutiae sets. Most minutiae-based methods suffer from several shortcomings. For example, extracting minutiae from a poor-quality fingerprint image may result in low matching accuracy. In addition, these methods may not fully utilize the rich discriminatory information available in the fingerprints with high computational complexity [17].
The image-based methods [1], [11], [12], [22], [27], [29], however, use features other than characteristics of minutiae from the fingerprint ridge pattern, such as local orientation and frequency, ridge shape, and texture information. The features for these methods may be extracted more reliably than those of minutiae. They usually require less preprocessing effort than minutiae-based methods using global information from a fingerprint, but they have limited ability to track variations in position, scale, and rotation angle of a fingerprint [27]. Invariance to an affine transform should be included for matching in order to deal with different input conditions and hence to enhance matching accuracy. Hybrid methods [2], [18], [19], [21] using features from both approaches have recently been researched. These methods have mostly the same problems as the minutiae-based methods.
One fine method uses a reference point with some image-based features [11] for fingerprint matching. In this method, the variation in position is canceled by registering the images with respect to a reference point that can be consistently detected in different instances of the same fingerprint. However, this approach exhibits deficiencies of typical image-based methods, which will be discussed in next section.
In this paper, an image-based algorithm using tessellated invariant moment features for fingerprint verification with a reference point is proposed to perform more accurately and in less processing time. It reliably finds a reference point with the proposed preprocessing and uses an alignment of the input image with the template. It also uses invariant moment features that are invariant to an affine transform to remedy the problems occurred in previous researches, and performs matching.
A fingerprint image is preprocessed to enhance the image by short time Fourier transform (STFT) analysis [3]. The STFT can be used to analyze the fingerprint image both in space and in frequency, helping to eliminate multi-spectral noise in the image. The algorithm simultaneously estimates all the intrinsic properties of the fingerprints, such as foreground region mask, local ridge orientation, and local ridge frequency, and uses these properties to enhance the fingerprint image.
A reference point is used to align a template and an input fingerprints before applying the local structure for verification. In the proposed algorithm, a global structure, which represents the maximum curvature in an orientation field image, is used to determine a unique reference point for all types of fingerprints including partial fingerprints. The position of the reference point is determined by the complex filtering methods [15], [20]. They find a unique point with the maximum curvature very successfully. The orientation of the reference point is determined by using the least mean square (LMS) orientation estimation algorithm [7], which estimates the orientation field using the gradient at each pixel and smoothes it with a Gaussian window.
An ROI centered on the reference point is then determined and tessellated into a predefined number of nonoverlapping square cells in order to minimize the effects of noise and nonlinear distortions. A set of fixed-length features consisting of seven invariant moments is extracted from each tessellated cell to represent the fingerprint as information of the local structure. Fingerprint verification is based on measures of similarity to the eigenvalue-weighted cosine (EWC) distance to match the two corresponding feature vectors of the input fingerprint image and template fingerprint image in the database. Experiments under various conditions have been done to evaluate the performance of the proposed method in terms of accuracy and computation speed, and to compare with other prominent methods using public databases.
The paper is organized as follows. 2 Prior related works and motivation, 3 Invariant moments are for the summary of prior related works and motivation, and a brief review of invariant moments, respectively. In Section 4, the proposed method is explained in detail and its experimental results are discussed in Section 5. Section 6 summarizes the conclusion.
Section snippets
Prior related works and motivation
Image-based methods are frequently used for fingerprint recognition. Among various image-based methods, Gabor feature-based methods [11], [22] present a relatively high matching accuracy by using a bank of Gabor filters to capture both the local and global features. By approximately making the methods rotation-invariant with multiple templates, however, these methods require significantly high processing times with large storage space, and performance degradation from the approximation.
Hybrid
Invariant moments
Moment features used in this paper can provide the properties of invariance to scale, position, and rotation [6]. We used moment analysis to extract invariant features from tessellated cells in an ROI. This section gives a brief description of the moment analysis.
For a 2-D continuous function f(x, y), the moment of order (p+q) is defined as:
A uniqueness theorem states that if f(x, y) is piecewise continuous and has nonzero values only in a finite part of
Proposed algorithm
The proposed algorithm contains five main steps, as shown in Fig. 1: preprocessing with STFT analysis, determination of reference point and image alignment, determination of ROI and its tessellation, invariant moment analysis, and matching with a similarity measure.
Experimental results
As public databases, FVC2000, FVC2002, and FVC2004 were established with the aim of providing a benchmark to determine the state-of-the-art techniques in fingerprint recognition applications. The fingerprints are well suited for testing a contemporary (online) fingerprint system and acquired by using modern capacitive and optical scanners.
Here the proposed algorithm was evaluated on fingerprint images taken from the FVC2002 database set_a,1 which contains four
Conclusion
We have proposed an enhanced image-based algorithm for fingerprint verification algorithm based on tessellated invariant moment features. The proposed algorithm performs the verification with five steps: fingerprint enhancement, reference point determination and image alignment, ROI determination and tessellation, invariant moment features extraction, and matching with EWC distance.
The proposed algorithm basically uses moment features invariant to scale, position and rotation to increase the
Acknowledgments
This research was financially supported by the Ministry of Commerce, Industry and Energy (MOCIE) and Korea Industrial Technology Foundation (KOTEF) through the Human Resource Training Project for Regional Innovation, and supported by the second stage of Brain Korea 21 Project.
Ju Cheng Yang received his B.S. in South-Central University for Nationalities, China and M.S. in Chonbuk National University, Korea. Now he is a Ph.D. candidate in Chonbuk National University, Korea. His research interests include image processing, pattern recognition, artificial intelligence, computer vision, and biometric. He is a student member of IEEE.
References (29)
- et al.
Fingerprint matching from minutiae texture maps
Pattern Recognit.
(2007) - et al.
Fingerprint enhancement using STFT analysis
Pattern Recognit.
(2007) - et al.
An efficient fingerprint verification system using integrated wavelet and Fourier-Mellin invariant transform
Image Vis. Comput.
(2004) An adaptive hybrid energy-based fingerprint matching technique
Image Vis. Comput.
(2005)- et al.
A hybrid wavelet-based fingerprint matcher
Pattern Recognit.
(2007) - et al.
Localization of corresponding points in fingerprints by complex filtering
Pattern Recognit. Lett.
(2003) - et al.
A hybrid fingerprint matcher
Pattern Recognit.
(2003) - et al.
How effective are landmarks and their geometry for face recognition
Comput. Vis. Image Underst.
(2006) - et al.
Fingerprint recognition using DCT features
Electron. Lett.
(2006) - et al.
Pattern Classification
(2000)
Pattern Classification
Digital Image Processing
Fingerprint image enhancement: algorithm and performance evaluation
IEEE Trans. Pattern Anal. Mach. Intell.
Generalizing discriminant analysis using the generalized singular value decomposition
IEEE Trans. Pattern Anal. Mach. Intell.
Cited by (80)
Fingerprints verification based on their spectrum
2016, NeurocomputingCitation Excerpt :Yang and Zhou [27] combined multiple enrolled impressions and used fusion scheme for combining the feature fusion and the decision fusion to improve the fingerprint verification systems. Yang and Park [28] presented a fingerprint verification algorithm using a set of fixed length moment features, and the invariant to the affine transform extracts from tessellated cells. The eigenvalue weighted cosine distance was used for evaluating the similarity between the input image and the template image in the database.
Unmatched minutiae: Useful information to boost fingerprint recognition
2016, NeurocomputingCitation Excerpt :Matching method is one of the most important components in fingerprint recognition. Current extensive techniques on matching method can be classified into four categories: minutiae based methods [3–11] in which minutiae such as ridge endings and ridge bifurcations in two fingerprints are extracted and matched, ridge based methods [12,13] in which the ridges of two fingerprints are extracted and matched, texture based methods [14,15] in which the local or global texture features of two fingerprints are extracted and matched, and neural network based methods [16,17] in which Artificial Neural Networks (ANNs) are introduced to match two fingerprints. Among these approaches, the minutiae based matching methods are the most popular methods for their robustness and reliability.
A Concise Survey on Biometric Recognition Methods
2023, International Journal of Computing and Digital SystemsA-Survey: Identification and Classification of Fingerprints via the Extreme Learning Machine Algorithm
2023, Inteligencia ArtificialFINGERPRINTS ANALYSIS USING AI ALGORITHM
2022, International Journal of Medical Toxicology and Legal MedicineFingerprint matching using rotational invariant orientation local binary pattern descriptor and machine learning techniques
2020, Cognitive Analytics: Concepts, Methodologies, Tools, and Applications
Ju Cheng Yang received his B.S. in South-Central University for Nationalities, China and M.S. in Chonbuk National University, Korea. Now he is a Ph.D. candidate in Chonbuk National University, Korea. His research interests include image processing, pattern recognition, artificial intelligence, computer vision, and biometric. He is a student member of IEEE.
Dong Sun Park is now a professor in Chonbuk National University, Korea. He received his B.S. in Korea University, Korea and M.S., Ph.D. degrees in the University of Missouri, USA. His research interests include image processing, pattern recognition, computer vision, and artificial intelligence. He has published lots of papers on the international conferences and journals. He is an association member of IEEE Computer Society.