Elsevier

Pattern Recognition

Volume 42, Issue 3, March 2009, Pages 452-464
Pattern Recognition

Integrating a wavelet based perspiration liveness check with fingerprint recognition

https://doi.org/10.1016/j.patcog.2008.06.012Get rights and content

Abstract

It has been shown that fingerprint scanners can be deceived very easily, using simple, inexpensive techniques. In this work, a countermeasure against such attacks is enhanced, that utilizes a wavelet based approach to detect liveness, integrated with the fingerprint matcher. Liveness is determined from perspiration changes along the fingerprint ridges, observed only in live people. The proposed algorithm was applied to a data set of approximately 58 live, 50 spoof and 28 cadaver fingerprint images captured at 0 and 2 s, from each of three different types of scanners, for normal conditions. The results demonstrate perfect separation of live and not live for the normal conditions. Without liveness module the commercially available verifinger matcher is shown to give equal error rate (EER) of 13.85% where false reject rate is calculated for genuine-live users and false accept rate is for genuine-not live, imposter-live and imposter-not live. The integrated system of fingerprint matcher and liveness module reduces EER to 0.03%. Results are also presented for moist and dry fingers simulated by glycerin and acetone, respectively. The system is further tested using gummy fingers and various deliberately simulated conditions including pressure change and adding moisture to the spoof to analyze the strength of the liveness algorithm.

Introduction

With more efficient communication techniques and more complex networked societies, human recognition has gained significance. Biometrics is defined as “automated methods for verifying or identifying the identity of a living individual based on physiological or behavioral characteristics” [1]. Biometrics has advantages over traditional identifiers like identity cards, signatures etc., in that they can not be forgotten, transferred, misplaced, duplicated, or stolen easily [1], [2]. Some examples of biometric identifiers include fingerprints, iris, hand geometry, voice, speech, face and gait. Fingerprints are the oldest, most studied and most widely used [3].

Fraudulent entry of an unauthorized person into a fingerprint recognition system by using faux fingerprint sample is termed spoofing [4]. Recently, different spoofing techniques have been reported, which include fake fingers using gelatin (gummy fingers), moldable plastic, clay, play-doh, wax and silicon, developed from casts of live fingers or latent fingerprints [4], [5], [6], [7], [8], [9]. Cadaver fingers have also been shown to reliably be scanned and verified by fingerprint devices of various technologies [4]. One efficient countermeasure against such a fraudulent attacks is liveness detection. As the name suggests, this technique checks whether the incoming biometric signal is coming from a live, genuine person [4], [9], [10].

Previously, liveness detection has been suggested as a countermeasure against spoofing of fingerprint scanners. Several liveness measures including pulse, pulse oximetry, electrocardiogram (ECG), temperature detection and multi-spectroscopy based techniques are suggested [5], [6], [7], [9], [10], [11], [12], [13], [14], [15], [16].

Liveness detection for fingerprint scanners can be applied using

  • Extra hardware to acquire liveness signs.

  • Soft processing techniques to extract liveness signs from already captured information.

Main drawbacks of the hardware based approaches are being bulky, expensive and vulnerable to the spoofing techniques as mentioned in the previous section. Few of the most accepted hardware methods are as follows:
  • Pulse oximetry is based on differential absorption of two wavelengths of light projected through the finger. While blood oxygen content is ignored, the pulse information is used for liveness detection. This method suffers with drawbacks including a requirement that the finger be completely covered for the test since ambient light may interfere [6].

  • Another suggested method is an ECG measurement. This method requires two contact points on the opposite sides of the subject's body to be able to perform the measurement. This method is bulky, and hence less practical [1], [4].

  • An elegant approach uses multi-spectral sensors to expose the finger to different optical wavelengths. This, as in others, requires extra hardware in a specially designed fingerprint scanner. A gelatin faux fingerprint is shown to have optical properties very similar to human skin [3].

  • Another method detects temperature of epidermis which typically is in the range 2530C. Main drawback of this method is vulnerability increases as the range of operating temperatures increases [6].

  • A US patent ‘anti-fraud biometric sensor that accurately detects blood flow’ by smarttouch LLC describes a method to determine finger blood-flow. The technique is based on pulse oximetry [17].

  • Another technique based on pulse oximetry is mentioned in Ref. [18], but the technique can be fooled using a translucent artificial fingerprint [4].

  • The electrical resistance of human skin with pre-specified range is used as liveness measure in Ref. [6]. This method can be spoofed by gelatin fingers as they have comparable moisture level as compared to human skin [5].

The proposed method, being purely software based, is cheaper and more flexible for future adaptations.

Previously, it has been shown that perspiration pores are present in the finger image [19], [20]. Fig. 1(a) shows an example pores and perspiration ducts from a histology of a finger. These pores have been used in Ref. [20] to improve fingerprint matching. Our research has also utilized perspiration arising from the pores as a measure of fingerprint liveness [8], [21].

Unlike cadaver or spoof fingers, live fingers demonstrate a distinctive spatial moisture pattern, when in physical contact with the capturing surface of the fingerprint scanner. As shown in Fig. 1(b), this pattern evolves in time due to the physiological phenomenon of perspiration. The pattern begins as ‘patchy’ areas of moisture around the pores spreading across the ridges over time. When considering the gray levels along the ridges, the pattern is sinusoidal in nature where peaks correspond to dark, moist areas and valleys correspond to light dry areas. Hence, we term this process as a ‘perspiration pattern’. Samples of live fingers compared to spoof are shown in Fig. 1(c). Previously, perspiration pattern based algorithms have been developed and depend on two fingerprint images captured over time. A signal processing based method has been developed, which formulates one-dimensional ridge signal, calculates statistical features and uses different classification methods like back propagation neural network, One R and discriminant analysis. This method is capable of producing classification rates in the range of 45–90% for several scanner technologies for approximately 30 live, 30 spoof and 14 cadaver fingerprint image pairs [4], [8]. By converting the fingerprint image into a one-dimensional signal of gray levels along the ridges, the disadvantage of this approach is that information that potentially characterizes liveness may be lost in this conversion.

This paper describes an approach which uses wavelet analysis of the entire fingerprint image to isolate the changing perspiration pattern. The algorithm is designed to specifically target liveness. The changing perspiration pattern can be considered as distinctive spatial property of the time series fingerprint images. These patterns mainly results from the physical surface properties, such as sweat pore positioning and pressure, roughness of a tactile fingerprint quality and so forth. The resultant effectiveness of the liveness algorithm depends heavily on efficient fingerprint texture characterization and extraction of the evolving pattern. Intuitively, wavelet filters were selected as they provide multiscale multiresolution framework. To confirm the choice of the filters, initially, simpler Fourier analysis was performed with virtually no success. Hence, further, sparse representation of wavelet coefficients was used to analyze perspiration phenomenon in time series fingerprint captures. After a pre-processing stage, wavelet analysis of the image is performed where wavelet packet analysis focuses on the high frequency and multiresolution analysis (MRA) focuses on low frequency components. A liveness measure is built by considering the changing wavelet coefficients. Preliminary work shows that the wavelet based algorithm gives a perfect classification for the same database for 0 and 5 s images [22], [23].

In this work, the wavelet based approach is enhanced and evaluated for a variety of factors. First, the time between images is reduced from 5 to 2 s (the minimum time possible by the available technology when data was collected). Second, liveness is integrated with a commercially available fingerprint matcher, verifinger [24]. One of the purposes of combining liveness with a matcher is for a simple quality check. It may be possible to spoof a liveness algorithm by submitting a poor quality image. By integrating with a matcher, baseline quality is guaranteed. In addition, it is important to consider liveness in the context of a system. Liveness is not a stand alone algorithm, but one which could have significant impact on the FRR even if it improves spoof FAR. Third, an extended data set collected from live subjects and a variety of fake fingers, prepared using play-doh, gelatin and cadaver fingers are used for the evaluation. Also, since perspiration is dependent on the environment, the designed system is tested for dry and wet fingerprint images. Last, to further analyze the strength of the algorithm, preliminary work is shown where deliberate attempts to spoof the perspiration liveness algorithm are performed, including change of pressure and simulation of perspiration by sprinkling water.

In the next two sections the complete system overview is presented along with the data management. In Section 4, the wavelet based liveness detection algorithm is described. The last two 5 Results, 6 Discussion and future work present results and conclusion.

Section snippets

System overview

This section has two subsections. The first subsection shows the flowchart of the different steps involved in the wavelet based liveness detection algorithm. The second subsection presents the integration of the liveness algorithm with fingerprint recognition.

Data management

Since the liveness algorithm relies on the evolving perspiration pattern, observed only in live people, a specific type of fingerprint data collected over known time intervals is required for evaluating the proposed system. Since no publicly available data sets have fingerprint time series captures, the data collected in our lab is used to test the algorithm. This data set is diverse as far as age, sex and ethnicity is concerned. Different age groups, ethnicities (Asian–Indian, Caucasian,

Wavelet based algorithm

The complete algorithm can be divided into three parts. The first part includes pre-processing steps to prepare the data for the wavelet analysis. The second part is the actual wavelet analysis. Post-processing steps to perform classification form the third part.

Results

This section shows the results generated after applying the algorithm to a data set of live, spoof, gummy and cadaver fingerprint images, described previously. Also, the results for different simulated conditions are also presented. This section is divided into three subsections.

The first subsection presents results of the liveness detection algorithm for the 2 s time window. The second subsection shows the effectiveness of embedding a liveness algorithm in a fingerprint recognition system. In

Discussion and future work

A new wavelet based method has been developed which is based on detecting perspiration pattern from a time series of fingerprint images measured directly from scanner. This method calculates wavelet coefficients on two images and characterizes perspiration changes in live fingers, thus separating them from spoof or dismembered fingers. Wavelet packet and MRA that isolate the perspiration changes and the total energy, given in Eq. (2), are used as measures to decide liveness. Threshold values,

Conclusion

A wavelet approach to detect liveness associated with fingerprint scanners is integrated with the Verifinger matcher. The liveness algorithm is based on detection of a perspiration pattern from two successive fingerprints captured at zeroth second after placement and after 2 s. The designed system checks for liveness for pairs classified as a match by the fingerprint matcher. This ensures liveness is only considered for good quality images. The liveness algorithm combines the use of Daubechies

Acknowledgments

This research has been partially funded by grants from the Center for Identification Technology (CITeR), the National Science Foundation (#0325333), and the Department of Homeland Security.

About the Author—ADITYA ABHYANKAR received the BE degree in Electronics and Telecommunication Engineering from Pune University, India in 2001. He received the MS and Ph.D. degrees from Clarkson University, NY, USA in 2003 and 2006 respectively. He worked as a post-doctoral fellow at Clarkson University, NY, USA in the academic year 2006-07. He worked as a consultant for Biometrics LLC, WV, USA in 2007. Currently he works as a Professor at Computer Engineering Department of Vishvakarma Institute

References (28)

  • J.D. Woodward et al.

    Biometrics

    (2003)
  • A. Jain et al.

    Biometrics: Personal Identification in Networked Society

    (1999)
  • D. Maltonie et al.

    Handbook of Fingerprint Recognition

    (2003)
  • S. Schuckers, Spoofing and anti-spoofing measures, Information Security Technical Report, vol. 7, 2002, pp....
  • T. Matsumoto, H. Matsumoto, K. Yamada, S. Hoshino, Impact of artificial gummy fingers on fingerprint systems, in:...
  • T. van der Putte, J. Keuning, Biometrical fingerprint recognition: don’t get your fingers burned, in: Proceedings of...
  • D. Willis, M. Lee, Six biometric devices point the finger at security, Biometrics Under our Thumb, Network Computing,...
  • R. Derakshani et al.

    Determination of vitality from a non-invasive biomedical measurement for use in fingerprint scanners

    Pattern Recognition

    (2003)
  • L. Thalheim, J. Krissler, Body check: biometric access protection devices and their programs put to the test, c’t...
  • V. Valencia, C. Horn, Biometric liveness testing, in: Biometrics, McGraw–Hill, New York,...
  • D. Osten, H. Carim, M. Arneson, B. Blan, Biometrics, personal authentication system, US Patent #5719950, February...
  • K. Seifried, Biometrics—what you need to know, in: Security portal, Available from:...
  • P. Lapsley, J. Less, D. Pare, N. Hoffman, Anti-fraud biometric sensor that accurately detects blood flow, in: US Patent...
  • P. Kallo, I. Kiss, A. Podmaniczky, J. Talosi, Detector for recognizing the living character of a finger in a...
  • Cited by (0)

    About the Author—ADITYA ABHYANKAR received the BE degree in Electronics and Telecommunication Engineering from Pune University, India in 2001. He received the MS and Ph.D. degrees from Clarkson University, NY, USA in 2003 and 2006 respectively. He worked as a post-doctoral fellow at Clarkson University, NY, USA in the academic year 2006-07. He worked as a consultant for Biometrics LLC, WV, USA in 2007. Currently he works as a Professor at Computer Engineering Department of Vishvakarma Institute of Information Technology, Pune. He also works as ’Research Associate’ at Clarkson University, NY, USA. He is also involved in consultancy with number of private industries. His research interests include signal and image processing, pattern recognition, wavelet analysis, biometric systems and bioinformatics.

    About the Author—STEPHANIE A.C. SCHUCKERS received the M.S. and Ph.D. degrees in Electrical Engineering from the University of Michigan, Ann Arbor, in 1994 and 1997, respectively, where she was a Whitaker Foundation Graduate Fellow. Currently, she is an Associate Professor with the Department of Electrical and Computer Engineering, Clarkson University, Potsdam, NY. Her primary research interest lies in the application of modern digital signal processing and pattern recognition to biomedical signals. Signals include the ECG, biometric signals like fingerprints, pulse oximetry, respiration, and electrocephalograms. Her work is funded by various sources, including National Science Foundation, American Heart Association, National Institute of Health and private industry.

    View full text