Advanced methods for two-class pattern recognition problem formulation for minutiae-based fingerprint verification
Introduction
Automatic fingerprint matching may be broadly classified as being either minutiae-based, correlation-based or image-based (Maio and Nanni, 2005, Lumini and Nanni, 2006, Nanni and Lumini, 2006) (for a good survey see Maio et al., 2003). Minutiae-based systems work on the minutiae-based representation of the fingerprint. Image-based approaches usually extract the features directly from the grey-level fingerprint image; then, the decision is made using these features. The performance of image-based techniques can be largely affected by non-linear distortions and noise present in the image; therefore, in general, minutiae-based techniques perform better than image-based ones. Anyway, image-based approaches may be the only viable choice, for instance, when image quality is too low to allow reliable minutia extraction.
Fingerprint matching based on minutia features is a well-researched problem (Tico and Kuosmanen, 2003, Chen et al., 2005). During the last four decades, various algorithms have been proposed to match two minutiae templates. A traditional way to calculate the similarity score for a minutiae-based system is n2/(NI × NR): where NI and NR represent the numbers of minutiae on the query and on the reference fingerprints, and n is the number of matched minutiae on both the fingerprints. In (Bazen and Gerez, 2003), the authors claim that using 2 × n/(NI + NR) to compute the similarity scores gives better results. Other approaches adopt heuristic methods to calculate the scores. For example, in (Jea and Govindaraju, 2005), the authors propose to use the number of matched minutiae, the numbers of minutiae points on the overlapping areas, and the average feature distances (the feature distances are obtained for each minutia considering the two nearest-neighbours) to calculate reliable similarity scores.
To our knowledge the only two works that try to solve the problem of minutiae-matching as a pattern recognition problem are Jea and Govindaraju, 2005, Jia et al., 2007. In (Jea and Govindaraju, 2005), the classification is performed by means of a neural networks trained by the following features: a similarity score calculated by an heuristic method, the number of matched minutiae n; the number of minutiae on query and reference fingerprints (NI, NR), and the above cited formulas for similarity calculation (n2/(NI × NR); 2 × n/(NI + NR)). In (Jia et al., 2007), the traditional minutiae-based matching task is studied as a classification task by using support vector machine (SVM). The SVM is trained on five features: the number of the minutiae in the two fingerprints; the number of mated minutiae; a feature based on the distance between the mated minutiae and the distance between each minutiae and its nearest neighbour minutiae; the score of a “standard” method to compare fingerprints.
In this work, we present a new method for minutiae-based fingerprint verification that approaches the problem as a two-class pattern recognition problem, where the unknown pattern is the response of a matching between the fingerprint of an unknown user and a stored one. With respect to previous works in the literature (i.e. Jea and Govindaraju, 2005, Jia et al., 2007), we propose several new features to classify the match into one of the two classes (“genuine” and “impostor”), moreover our experiments show that our method could be coupled with any existing method to calculate the similarity scores. In order to improve the verification accuracy, we generate an higher dimensional feature set, obtained by several Fisher transform projections (Belhumeur et al., 1997), and only the most discriminative features are selected by the sequential forward floating selection (Pudil et al., 1994) feature selection algorithm to be used for the classification.
We show that coupling the features extracted from the matched minutiae and the score obtained by a commercial matcher (whose details can be even unknown) we obtain an equal error rate (EER) and a zero false acceptation rate (ZeroFAR) lower than that obtained using only the commercial matcher. Such performance can be further improved by adopting the set of “artificially combined” features.
Section snippets
Our approach
Our system works on the minutiae-based representation of the fingerprint. The four databases collected for FVC2002 (bias.csr.unibo.it/fvc2002/) are used for the experiments. The minutiae extraction and matching has been performed by the commercial system Biometrika for DB2 (since the Biometrika software works only using the images of the database DB2 which have been acquired by its own sensor), and by CUBS fingerprint toolbox1 minutiae extractor coupled with Tico’s matcher (
Experiments
Each algorithm has been tested using the FVC2002 testing protocol (as in other papers e.g. Lumini and Nanni, 2007, Maio and Nanni, 2006), by performing the following matching attempts:
- •
Genuine recognition attempts: the template of each impression is matched against the remaining impressions of the same finger, but avoiding symmetric matches (i.e. if the template of impression j is matched against impression k, template k is not matched against impression j);
- •
Impostor recognition attempts: the
Conclusions
The main contribution of this work is the idea of handling a minutiae-based fingerprint verification problem as a two-class pattern recognition problem. A feature vector obtained by minutiae extraction and matching is classified into one of the two classes (genuine or impostor) by support vector machines. We studied the effects on the performance of feature selection methods and “artificial” feature generation. There are three elements for a good matcher: a precise feature extraction, a strong
Acknowledgement
This work has been supported by Italian PRIN prot. 2004098034 and by European Commission IST-2002-507634 Biosecure NoE projects.
References (17)
- et al.
Fingerprint matching by thin-plate spline modeling of elastic deformations
Pattern Recognition
(2003) - et al.
A minutiae-based partial fingerprint recognition system
Pattern Recognition
(2005) - et al.
Fingerprint matching based on weighting method and the SVM
Neurocomputing
(2007) - et al.
Two-class fingerprint matcher
Pattern Recognition
(2006) - et al.
An efficient fingerprint verification system using integrated Gabor filters and Parzen Window Classifier
Neurocomputing
(2005) - et al.
Fingerprint verification by fusion of optical and capacitive sensors
Pattern Recognition Lett.
(2004) - et al.
Floating search methods in feature selection
Pattern Recognition Lett.
(1994) - et al.
Eigenfaces vs. Fisherfaces: recognition using class specific linear projection
IEEE Trans. Pattern Anal. Machine Intell.
(1997)
Cited by (13)
A fingerprint and finger-vein based cancelable multi-biometric system
2018, Pattern RecognitionCitation Excerpt :The performance of this image-based multi-biometric system can be seriously affected by non-linear distortion and noise present in the fingerprint image. Therefore, in general, for the fingerprint image which has a set of salient points, minutia-based techniques are preferred to image-based ones [6]. Different to [4], in [5], minutia-based features are first extracted from fingerprint and finger-vein images, and then the fused feature point-sets from template and query images are matched using a point pattern matching and weight matching algorithm.
Fingerprints verification based on their spectrum
2016, NeurocomputingCitation Excerpt :Nanni and Lumini [21] presented another approach similar to [18] but the feature vector is projected onto a one dimensional Karhunen Loeve space and then classified into one of two fingerprint classes (genuine or impostor). Also, they used minutiae based features with Support Vector Machines [22]; the results show a remarkable performance improvement. Also, they presented and compared between the image based and minutiae based fingerprint matching in [23].
FEATURES EXTRACTION OF FINGERPRINTS BASED BAT ALGORITHMS
2022, International Journal on Technical and Physical Problems of EngineeringHandbook of fingerprint recognition: Third edition
2022, Handbook of Fingerprint Recognition: Third EditionIMPROVEMENT of the PERFORMANCE of FINGERPRINT VERIFICATION USING A COMBINATORIAL APPROACH
2018, Biomedical Engineering - Applications, Basis and CommunicationsNovel approaches of biometric finger print minutiae detection and extraction process
2017, International Journal of Mechanical Engineering and Technology