25-03-2011, 10:59 AM
[attachment=10962]
Abstract
This paper investigates correlation-based fingerprint authentication schemes that can be used for mobile devices. The investigated algorithms were implemented with a J2ME
Environment on the application layer In order to reduce the resources demanded for the mobile device environment, we also propose a new hierarchical correlation-based scheme based on the idea that the overall authentication can be decomposed into partial autocorrelations. The algorithms have been tested on a J2ME CDC 1.0 emulator of a smart mobile phone.
1. Introduction
Today, a mobile phone can be integrated with a camera, a GPRS, a radio, a MP3 player, a web browser and even a TV. It is foreseeable that future mobile devices will just be
More powerful and function like hand held computers. With this trend of convergence, potential security problems have become more threatening and harmful. This
Urges stronger protections against data leaking and illegitimate use of the device. Biometric authentication can ensure genuine user presence, thus enhancing the privacy protection.
Only recently, a few products of biometric-enabled mobile devices have been announced available to consumers. However, different manufacturers tend to have their own standards and proprietary technology. In most current commercial solutions, the biometric function is embedded in the system hardware and is expensive.
We consider to deploy biometric authentication in the application layer so that better extendability and portability can be achieved for general mobile devices.
Our application is developed using Java 2 Micro Edition (J2ME) [2]. J2ME is a green version of Java. It inherits Java's main benefit of being platform independent as well as
object oriented. Moreover, J2ME was especially designed to fit resource-constrained embedded systems. Its applications can be emulated on a PC during the development stage and then easily uploaded to PDAs or mobile phones, with out the need of expensive system-specific kits and hardware.
J2ME applications should be designed to consume as little
resource as possible. To meet this special requirement, we develop a new hierarchical correlation algorithm for fingerprint authentication on mobile devices.The proposed image correlation. To investigate the authentication performance, a worst case scenario for the correlation-based algorithms was considered where fingerprints with plastic distortions are used for testing in our experiments.
Hierarchical Fingerprint Authentication:
Most existing algorithms for fingerprint matching are based on ridge endings and bifurcations (minutiae) [5]. In those schemes, authentication is approved only if the number of matched minutiae exceeds a predefined threshold. For mobile devices, the fingerprint sensor is usually quite small. Hence, partial and non-overlapping fingerprints are often obtained. This tends to reduce the performance of a minutiae-based fingerprint matching approach. Moreover, minutiae-based algorithms often require a few intermediate image processing steps such as orientation extraction [7, 8] and ridge thinning [1], which will increase the complexity of the J2ME application on mobile devices.
The correlation-based fingerprint matching uses overall inform ation-piedi fingerprint image. A synthetic information provided in a fingerprint image. A synthetic filter is often built as a template using a number of training examples [3]. When a test fingerprint perfectly matches with the filter (template), a well-defined peak will appear in the resulting correlation plane. Otherwise, a flat correlation
output is expected to be observed.
Minimum average correlation energy (MACE) filter The MACE filter was designed to suppress the sidelobes of correlation plane such that a sharp correlation peak
can be produced. Assuming N training images of a subject, each image has a total of d pixels. For the i'th training image, the columns of its 2D Fourier transform is concatenated to form a column vector xi containing d elements. A matrix X from N training images is then defined as
X = [X1,X2 * XN]T.
The 2D MACE filter obtained in the frequency domain is also ordered in a column vector h. The i'th correlation output at the origin is constrained to a prespecified value ui, which can be represented as
where the superscript '+' denotes a conjugate transpose. Note that c(O) is also referred to the correlation peak value. On the other hand, based on Parseval's theorem, the aver-
age of the correlation plane energies, Eave, can be obtained directly from the frequency domain by
where the superscript '*' denotes complex conjugation and D is a diagonal matrix of size d x d whose diagonal elements are the power spectrum of xi Minimizing the average correlation energy Eave subjecting to the constraints placed in (1) leads to the MACE filter solution
Hierarchical correlation-based authentication :-
Conventional correlation-based authentications use fullsized fingerprint images. It has been reported that down-
smln Odimaeto26x56pesrsusinbtr sampling 5OOdpi images to 256 x 256 pixels results in better performance compared to other resolutions [6]. However for mobile devices, this still consumes too much memory and computing power. Therefore, we consider to use partial images at each time of correlation computation. Let us first consider a simple ID case. In the space domain, correlation of r [k] with a target t [k] leads to the following correlation output
The above evaluation can be easily extended for 2D cases. It clearly shows that for autocorrelation, the output peak at the origin is equal to the sum of peak values obtained from the corresponding fractions of the original segment. If the fractions are from other sources, the difference between the peak sum and the original peak value from the target source will not be zero. Based on this idea, we propose a correlation-based hierarchical fingerprint authentication scheme as shown in Figure 2.
The key modules in Figure 2 are described as follows. In the enrollment stage, a template is constructed (possibly offline) from a set of training images based on the MACE
filter design as described previously. The template is represented in the space domain and will be stored in the mobile device.
In the authentication stage, three donut rings will be first extracted from the test fingerprint's core center by defining three concentric circles. For example as shown in Figure 3, the inner donut ring R1 is defined by concentric circles."Ciand er donut ring R, is defined by C1 and C2. The outer donut ring R2 iS defined by C2 and C3. The overall donut ring R3 is defined by Ci and C3. Corresponding parts in the template will also be extracted using concentric circles with the same diameters, namely T1, T2 and T3. The donut rings R1, R2 and R3 from the test fingerprint are then correlated with their corresponding template parts T1, T2 and T3 respectively, yielding three correlation peak values Pi, P2 and p3.
Abstract
This paper investigates correlation-based fingerprint authentication schemes that can be used for mobile devices. The investigated algorithms were implemented with a J2ME
Environment on the application layer In order to reduce the resources demanded for the mobile device environment, we also propose a new hierarchical correlation-based scheme based on the idea that the overall authentication can be decomposed into partial autocorrelations. The algorithms have been tested on a J2ME CDC 1.0 emulator of a smart mobile phone.
1. Introduction
Today, a mobile phone can be integrated with a camera, a GPRS, a radio, a MP3 player, a web browser and even a TV. It is foreseeable that future mobile devices will just be
More powerful and function like hand held computers. With this trend of convergence, potential security problems have become more threatening and harmful. This
Urges stronger protections against data leaking and illegitimate use of the device. Biometric authentication can ensure genuine user presence, thus enhancing the privacy protection.
Only recently, a few products of biometric-enabled mobile devices have been announced available to consumers. However, different manufacturers tend to have their own standards and proprietary technology. In most current commercial solutions, the biometric function is embedded in the system hardware and is expensive.
We consider to deploy biometric authentication in the application layer so that better extendability and portability can be achieved for general mobile devices.
Our application is developed using Java 2 Micro Edition (J2ME) [2]. J2ME is a green version of Java. It inherits Java's main benefit of being platform independent as well as
object oriented. Moreover, J2ME was especially designed to fit resource-constrained embedded systems. Its applications can be emulated on a PC during the development stage and then easily uploaded to PDAs or mobile phones, with out the need of expensive system-specific kits and hardware.
J2ME applications should be designed to consume as little
resource as possible. To meet this special requirement, we develop a new hierarchical correlation algorithm for fingerprint authentication on mobile devices.The proposed image correlation. To investigate the authentication performance, a worst case scenario for the correlation-based algorithms was considered where fingerprints with plastic distortions are used for testing in our experiments.
Hierarchical Fingerprint Authentication:
Most existing algorithms for fingerprint matching are based on ridge endings and bifurcations (minutiae) [5]. In those schemes, authentication is approved only if the number of matched minutiae exceeds a predefined threshold. For mobile devices, the fingerprint sensor is usually quite small. Hence, partial and non-overlapping fingerprints are often obtained. This tends to reduce the performance of a minutiae-based fingerprint matching approach. Moreover, minutiae-based algorithms often require a few intermediate image processing steps such as orientation extraction [7, 8] and ridge thinning [1], which will increase the complexity of the J2ME application on mobile devices.
The correlation-based fingerprint matching uses overall inform ation-piedi fingerprint image. A synthetic information provided in a fingerprint image. A synthetic filter is often built as a template using a number of training examples [3]. When a test fingerprint perfectly matches with the filter (template), a well-defined peak will appear in the resulting correlation plane. Otherwise, a flat correlation
output is expected to be observed.
Minimum average correlation energy (MACE) filter The MACE filter was designed to suppress the sidelobes of correlation plane such that a sharp correlation peak
can be produced. Assuming N training images of a subject, each image has a total of d pixels. For the i'th training image, the columns of its 2D Fourier transform is concatenated to form a column vector xi containing d elements. A matrix X from N training images is then defined as
X = [X1,X2 * XN]T.
The 2D MACE filter obtained in the frequency domain is also ordered in a column vector h. The i'th correlation output at the origin is constrained to a prespecified value ui, which can be represented as
where the superscript '+' denotes a conjugate transpose. Note that c(O) is also referred to the correlation peak value. On the other hand, based on Parseval's theorem, the aver-
age of the correlation plane energies, Eave, can be obtained directly from the frequency domain by
where the superscript '*' denotes complex conjugation and D is a diagonal matrix of size d x d whose diagonal elements are the power spectrum of xi Minimizing the average correlation energy Eave subjecting to the constraints placed in (1) leads to the MACE filter solution
Hierarchical correlation-based authentication :-
Conventional correlation-based authentications use fullsized fingerprint images. It has been reported that down-
smln Odimaeto26x56pesrsusinbtr sampling 5OOdpi images to 256 x 256 pixels results in better performance compared to other resolutions [6]. However for mobile devices, this still consumes too much memory and computing power. Therefore, we consider to use partial images at each time of correlation computation. Let us first consider a simple ID case. In the space domain, correlation of r [k] with a target t [k] leads to the following correlation output
The above evaluation can be easily extended for 2D cases. It clearly shows that for autocorrelation, the output peak at the origin is equal to the sum of peak values obtained from the corresponding fractions of the original segment. If the fractions are from other sources, the difference between the peak sum and the original peak value from the target source will not be zero. Based on this idea, we propose a correlation-based hierarchical fingerprint authentication scheme as shown in Figure 2.
The key modules in Figure 2 are described as follows. In the enrollment stage, a template is constructed (possibly offline) from a set of training images based on the MACE
filter design as described previously. The template is represented in the space domain and will be stored in the mobile device.
In the authentication stage, three donut rings will be first extracted from the test fingerprint's core center by defining three concentric circles. For example as shown in Figure 3, the inner donut ring R1 is defined by concentric circles."Ciand er donut ring R, is defined by C1 and C2. The outer donut ring R2 iS defined by C2 and C3. The overall donut ring R3 is defined by Ci and C3. Corresponding parts in the template will also be extracted using concentric circles with the same diameters, namely T1, T2 and T3. The donut rings R1, R2 and R3 from the test fingerprint are then correlated with their corresponding template parts T1, T2 and T3 respectively, yielding three correlation peak values Pi, P2 and p3.
