DETECTION AND RECTIFICATION OF DISTORTED FINGERPRINTS

DETECTION AND RECTIFICATION

OF DISTORTED FINGERPRINTS

Abstract—Elastic distortion of fingerprints is one of the major causes for false non-match. While this problem affects all fingerprint recognition applications, it is especially dangerous in negative recognition applications, such as watchlist and deduplication applications. In such applications, malicious users may purposely distort their fingerprints to evade identification. In this paper, we proposed novel algorithms to detect and rectify skin distortion based on a single fingerprint image. Distortion detection is viewed as a two-class classification problem, for which the registered ridge orientation map and period map of a fingerprint are used as the feature vector and a SVM classifier is trained to perform the classification task. Distortion rectification (or equivalently distortion field estimation) is viewed as a regression problem, where the input is a distorted fingerprint and the output is the distortion field. To solve this problem, a database (called reference database) of various distorted reference fingerprints and corresponding distortion fields is built in the offline stage, and then in the online stage, the nearest neighbor of the input fingerprint is found in the reference database and the corresponding distortion field is used to transform the input fingerprint into a normal one. Promising results have been obtained on three databases containing many distorted fingerprints, namely FVC2004 DB1, Tsinghua Distorted Fingerprint database, and the NIST SD27 latent fingerprint database.

EXISTING SYSTEM:

Due to the vital importance of recognizing distorted fingerprints,  researchers have proposed a number of methods

which can be coarsely classified into four categories It is desirable to automatically detect distortion during fingerprint acquisition so that severely distorted fingerprints can be rejected. Several researchers have proposed to detect improper force using specially designed hardware .WE  proposed to detect excessive force and torque exerted by using a force sensor. They showed that controlled fingerprint acquisition leads to improved matching performance . In existing  proposed to detect distortion by detecting deformation of a transparent film attached to the sensor surface. Dorai et al.  proposed to detect distortion by analyzing the motion in video of fingerprint The most popular way to handle distortion is to make the fingerprint matcher tolerant to distortion . In other words, they deal with distortion on a case by case basis, i.e., for every pair of fingerprints to be compared. For the most widely used minutiae-based fingerprint matching method, the following three types of strategies have been adopted to handle distortion: (i) assume a global rigid transformation and use a tolerant box of fixed size  or adaptive size  to compensate for distortion; (ii) explicitly model the spatial transformation by thin plate spline (TPS) model ; and (iii) enforce constraint on  istortion locally. Various methods for handling distortion during matching have also been used in image-based matcher  or skeleton-based matcher

PROPOSED SYSTEM:

In this paper, novel algorithms are proposed to deal with the fingerprint distortion problem. Given an input fingerprint, distortion detection is performed first. If it is determined to be distorted, distortion rectification is performed to transform the input fingerprint into a normal one. A distorted fingerprint is analogous to a face with expression, which affects the matching accuracy of face recognition systems. Rectifying a distorted fingerprint into a normal fingerprint is analogous to transforming a face with expression into a neutral face, which can improve face

recognition performance. In this paper, distortion detection is viewed as a two class classification problem, for which the registered ridge orientation map and period map of a fingerprint are used as the feature vector and a SVM classifier is trained to perform the classification task. Distortion rectification (or equivalently distortion field estimation) is viewed as a regression problem, where the input is a distorted fingerprint and the output is the distortion field. To solve this problem, a database of various distorted reference fingerprints and corresponding distortion fields is built in the offline stage, and then in the online stage, the nearest neighbor of the input fingerprint is found in the database of distorted reference fingerprints and the corresponding distortion field is used to rectify the input fingerprint. An important property of the proposed system is that it does not require any changes to existing fingerprint sensors and fingerprint acquisition procedures. Such property is important for convenient incorporation into existing fingerprint recognition systems. The proposed system has been evaluated on three databases, FVC2004 DB1 whose images are markedly affected by distortion, Tsinghua distorted fingerprint database

which contains 320 distorted fingerprint video files, and NIST SD27 latent fingerprint database. Experimental results demonstrate that the proposed algorithms can improve the matching accuracy of distorted fingerprints evidently.

Module 1

Fingerprint distortion detection

Fingerprint distortion detection can be viewed as a two class classification problem. We used the registered ridge

orientation map and period map as the feature vector, which is classified by a SVM classifier.

Fingerprint Registration In order to extract meaningful feature vector, fingerprints have to be registered in a fixed coordinate system. For this task, we propose a multi-reference based fingerprint registration approach. In the following, we describe how the reference fingerprints are prepared in the offline stage, and how to register an input fingerprint in the online stage.

Reference Fingerprints In order to learn statistics of realist fingerprint distortion, we collected a distorted fingerprint database called Tsinghua distorted fingerprint database. A FTIR fingerprint scanner with video capture functionality was used for data collection. Each participant is asked to press a finger on the scanner in a normal way, and then distort the finger by applying a lateral force or a torque and gradually increase the force. Online Fingerprint Registration In the online stage, given an input fingerprint, we perform the registration w.r.t. registered reference fingerprints. Level 1 features (orientation map, singular points, period map) are extracted using traditional algorithms. According to whether the upper core point is detected or not, the registration approach can be classified into two cases. If the upper core point is not detected, we do a full search to find the pose information, namely solve the following optimization formula: arg max x;y;a;i kOrientDiffðROiðx; y; aÞ;OÞ _ utk0; (1) where x and y denote the translation parameters, a denotes the rotation parameter, i denotes the corresponding reference fingerprint ID, O is the orientation map of the input fingerprint, ROi denotes the orientation map of the ith reference

fingerprint, function OrientDiffðÞ computes the difference of two orientation maps at each location, k _ k0 counts the number of nonzero elements, and ut is the threshold, which is empirically set as 10 degrees. Note that the ridge orientation map is defined on blocks of 16 _ 16 pixels. Finally, we register the ridge orientation map and period map of the input fingerprint to the fixed coordinate system by using the obtained pose information.  

Module 2

Statistical Modeling of Distortion Fields

In order to learn statistical fingerprint distortion model, we need to know the distortion fields (or deformation fields) between paired fingerprints (the first frame and the last frame of each video) in the training set. The distortion field between a pair of fingerprints can be estimated based on the corresponding minutiae of the two fingerprints. Unfortunately, due to the severe distortion between paired fingerprints, existing minutiae matchers cannot find corresponding minutiae reliably. Thus, we extract minutiae in the first frame using VeriFinger and perform minutiae tracking in each video. Since the relative motion between adjacent frames is small, reliable minutiae correspondences between the first frame and the last frame can be found by this method. Given the matching minutiae of a pair of fingerprints, we estimate the transformation using thin plate spline model  We define a regular sampling grid on the normal fingerprint and compute the corresponding grid (called distortion grid) on the distorted fingerprint using the TPS model.

Module 3

Distorted fingerprint rectification

A distorted fingerprint can be thought of being generated by applying an unknown distortion field d to the normal fingerprint, which is also unknown. If we can estimate the distortion field d from the given distorted fingerprint, we can easily rectify it into the normal fingerprint by applying the inverse of d. So we need to address a regression problem, which is quite difficult because of the high dimensionality of the distortion field (even if we use a block-wise distortion field). In this paper, a nearest neighbor regression approach is used for this task. The proposed distorted fingerprint rectification algorithm consists of an offline stage and an online stage. In the offline stage, a database of distorted reference fingerprints is generated by transforming several normal reference fingerprints with various distortion fields sampled from the statistical model of distortion fields. In the online stage, given a distorted input fingerprint (which is detected by the algorithm described in Section 3), we retrieval its nearest neighbor in the distorted reference fingerprint database and then use the inverse of the corresponding distortion field to rectify the distorted input fingerprint.

Module 4

Distorted Reference Fingerprint Database

To generate the database of distorted reference fingerprints, we use nref ¼ 100 normal fingerprints from FVC2002 DB1. The distortion fields are generated by uniformly sampling the subspace spanned by the first two principle components. For each basis, 11 points are uniformly sampled in the interval. See Fig. 9 for an example of generating distortion fields and applying such distortion fields to a reference fingerprint to generate corresponding distorted fingerprints. For visualization purpose, only one reference fingerprint (the fingerprint located at the origin of the coordinate system) is used to generate the database of distorted reference fingerprints, and for each basis, five

points are sampled. In practice, multiple reference fingerprints are used to achieve better performance. Also note that instead of storing the fingerprint image, we store the ridge orientation map and period map of each fingerprint in the reference database.

Module 5

Distortion Field Estimation by Nearest Neighbor Search

Distortion field estimation is equal to finding the nearest neighbor among all distorted reference fingerprints.  The similarity is measured based on level 1 features of fingerprint, namely ridge orientation map and period map. We conjecture that distortion detection and rectification of human experts also relies on these features instead of minutiae. The similarity computation method is different depending on whether the upper core point can be detected in the input fingerprint. If the upper core point is detected, we translate the input fingerprint by aligning the upper core point to center point. Then we do a full search of u in the interval ½_30; 30_ for the maximum similarity. For a specific u, the similarity between two fingerprints is computed as follows:  s ¼ sO1 þ sO2 m _wO 1 sO 1 þ wO 2 sO 2 _ þ sP1 þ sP2 m _wP1 sP1 þ wP2 sP2  where m denotes the number of blocks in the overlapping area, sO1 and sO2 denote the number of blocks with similar orientation above and below the center point, sP1 and sP2 denote the number of blocks with similar period above and below the center point, and the four weights wO1, wO2, wP1, wP2, are empirically set as 1, 0.5, 1, 1.5, respectively.

CONCLUSION

False non-match rates of fingerprint matchers are very high in the case of severely distorted fingerprints. This generates a security hole in automatic fingerprint recognition systems which can be utilized by criminals and terrorists. For this reason, it is necessary to develop a fingerprint distortion detection and rectification algorithms to fill the hole. This paper described a novel distorted fingerprint detection and rectification algorithm. For distortion detection, the registered ridge orientation map and period map of a fingerprint are used as the feature vector and a SVM classifier is trained to classify the input fingerprint as distorted or normal. For distortion rectification (or equivalently distortion field estimation), a nearest neighbor regression approach is used to predict the distortion field from the input distorted fingerprint and then the inverse of the distortion field is used to transform the distorted fingerprint into a normal one. The experimental results on FVC2004 DB1, Tsinghua DF database, and NIST SD27 database showed that the proposed algorithm can improve recognition rate of distorted fingerprints evidently A major limitation of the current approach is efficiency. Both detection and rectification steps can be significantly speeded up if a robust and accurate fingerprint registration algorithm can be developed. Another limitation is that the

current approach does not support rolled fingerprints. It is difficult to collect many rolled fingerprints with various distortion types and meanwhile obtain accurate distortion fields for learning statistical distortion model. It is our ongoing work to address the above limitations

REFERENCES

[1] X. Si, J. Feng, and J. Zhou, “Detecting fingerprint distortion from a single image,” in Proc. IEEE Int. Workshop Inf. Forensics Security, 2012, pp. 1–6.

 [2] D. Maltoni, D. Maio, A. K. Jain, and S. Prabhakar, Handbook of Fingerprint Recognition, 2nd ed. Berlin, Germany: Springer-Verlag, 2009.

[3] FVC2006: The fourth international fingerprint verification competition. (2006). [Online]. Available: http://bias.csr.unibo.it/fvc2006/

[4] V. N. Dvornychenko, and M. D. Garris, “Summary of NIST latent fingerprint testing workshop,” Nat. Inst. Standards Technol., Gaithersburg, MD, USA, Tech. Rep. NISTIR 7377, Nov. 2006.

[5] Neurotechnology Inc., VeriFinger. (2009). [Online]. Available: http://www.neurotechnology.com

[6] L. M. Wein and M. Baveja, “Using fingerprint image quality to improve the identification performance of the U.S. visitor and immigrant status indicator technology program,” Proc. Nat. Acad. Sci. USA, vol. 102, no. 21, pp. 7772–7775, 2005.

[7] S. Yoon, J. Feng, and A. K. Jain, “Altered fingerprints: Analysis and detection,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 34, no. 3, pp. 451–464, Mar. 2012.

[8] E. Tabassi, C. Wilson, and C. Watson, “Fingerprint image quality,” Nat. Inst. Standards Technol., Gaithersburg, MD, USA, Tech. Rep. NISTIR 7151, Aug. 2004.

[9] F. Alonso-Fernandez, J. Fi_errez-Aguilar, J. Ortega-Garcia, J. Gonzalez- Rodriguez, H. Fronthaler, K. Kollreider, and J. Big€un, “A comparative study of fingerprint image-quality estimation methods,” IEEE Trans. Inf. Forensics Security, vol. 2, no. 4, pp. 734–743, Dec. 2007.