Performance Evaluation of Fingerprint Biometrics Systems for e-Business Access Control

Abstract

The performance of a fingerprint authentication system in electronic business depends on the accurate analysis of the quality of input fingerprints. This paper investigates distorted and misaligned fingerprints caused by environmental noise such as oil, wrinkles, dry skin, dirt and displacement of the query fingerprint with the database fingerprint template during matching. The noisy, distorted and/or misaligned fingerprint produced as a 2-D on x-y image, is enhanced and optimized using a new hybrid modified Gabor filter-hierarchal structure check (MGF-HSC) system model based on an MGF integrated with a HSC. Our findings indicate that structural matching is experimentally confirmed as a reliable matching technique in a fingerprint authentication system, as distorted fingerprints can easily be matched, even when the pixel value detected is low, while the rotational and geometrical transformation that affects fingerprint authentication is addressed. The MGF-HSC approach minimizes false fingerprint matching and the dominant effect of distortion and misalignment of fingerprints to an acceptable level. Our approach, which is benchmarked with publicly available methods, deals with 60 % heavy distortions and misaligned fingerprint and exhibits reliable matching performance for promising e-business security work in terms of visual quality, false rejection rate and false acceptance rate.

ANNEXURE

Thus, parameter translation (PR) in Eq. (1),

$$ PR{}_{\left( \phi \right)} = \left\{ \begin{aligned} PR_{\left( \phi \right)} \,\,\,\,\,\,\,occurs,\,\,\,\,\,\,\,\,if\,\,\,\,\,\phi_{d1} = \phi_{d2} \hfill \\ \hfill \\ return\,\,\,\,\,\,failure,\,\,\,\,\,\,\,\,if\,\,\,\,\,\phi_{d1} \ne \phi_{d2} \hfill \\ \end{aligned} \right. $$

(1)

For $x$-component in Eq. (2),

$$ PR{}_{\left( x \right)} = \left\{ \begin{aligned} PR_{\left( x \right)} \,\,\,\,\,\,\,occurs,\,\,\,\,\,\,\,\,if\,\,\,\,\,x_{1} = x_{2} \hfill \\ \hfill \\ return\,\,\,\,\,\,failure,\,\,\,\,\,\,\,\,if\,\,\,\,\,x_{1} \ne x_{2} \hfill \\ \end{aligned} \right. $$

(2)

and for $y$-component in Eq. (3),

$$ PR{}_{\left( y \right)} = \left\{ \begin{aligned} PR_{\left( y \right)} \,\,\,\,\,\,\,occurs,\,\,\,\,\,\,\,\,if\,\,\,\,\,y_{1} = y_{2} \hfill \\ \hfill \\ return\,\,\,\,\,\,failure,\,\,\,\,\,\,\,\,if\,\,\,\,\,y_{1} \ne y_{2} \hfill \\ \end{aligned} \right. $$

(3)

For parameter distance in Eq. (4) $\left( {x,y} \right)$,

$$ {\text{m}} . {\text{ax}}\left( {{\text{MP}}_{ 1} - {\text{MP}}_{ 2} } \right) $$

(4)

The second distance is in (5), for $x$-component in Eq. (5),

$$ = m.ax\left( {MP_{2} - MP_{1} } \right) $$

(5)

Thus, parameter translation (PT) in Eq. (6),

$$ PT{}_{\left( x \right)} = \left\{ \begin{aligned} PT_{\left( x \right)} \,\,\,\,\,\,\,occurs,\,\,\,\,\,\,\,\,if\,\,\,\,\,x_{1} = x_{2} \hfill \\ \hfill \\ return\,\,\,\,\,\,failure,\,\,\,\,\,\,\,\,if\,\,\,\,\,x_{1} \ne x_{2} \hfill \\ \end{aligned} \right. $$

(6)

For the$y$-component in Eq. (7),

$$ PT{}_{\left( y \right)} = \left\{ \begin{aligned} PT_{\left( y \right)} \,\,\,\,\,\,\,occurs,\,\,\,\,\,\,\,\,if\,\,\,\,\,y_{1} = y_{2} \hfill \\ \hfill \\ return\,\,\,\,\,\,failure,\,\,\,\,\,\,\,\,if\,\,\,\,\,y_{1} \ne y_{2} \hfill \\ \end{aligned} \right. $$

(7)

For the $\phi$-component in Eq. (8)

$$ PT{}_{\left( \phi \right)} = \left\{ \begin{aligned} PT_{\left( \phi \right)} \,\,\,\,\,\,\,occurs,\,\,\,\,\,\,\,\,if\,\,\,\,\,\phi_{1} = \phi_{2} \hfill \\ \hfill \\ return\,\,\,\,\,\,failure,\,\,\,\,\,\,\,\,if\,\,\,\,\,\phi_{1} \ne \phi_{2} \hfill \\ \end{aligned} \right. $$

(8)

Condition 1: Matching with a predefined threshold value as in Eq. (9),

$$ M_{(m)} = \left\{ {\begin{array}{*{20}c} {Q \equiv P,} & {} & {if} & {\hbox{min} } & {utiae \ge threshold} \\ {} & {} & {} & {} & {} \\ {return} & {failure} & {otherwise} & {} & {} \\ \end{array} } \right. $$

(9)

Condition 2: Matching with a specified tolerance value as in Eq. (10),

$$ Match_{\,\,\,(Q,P)} = \left\{ {\begin{array}{*{20}c} {Q \equiv P,} & {} & {if} & {x \le \,\,Tolerance} \\ {} & {} & {} & {y \le \,\,Tolerance} \\ {} & {} & {} & {\theta \le \,\,\,Tolerance} \\ {} & {} & {} & {} \\ {return} & {failure} & {} & {otherwise} \\ \end{array} } \right. $$

(10)

$$ FRR = \frac{G}{N} $$

(11)

where $G$ the number of valid users fingerprint who are incorrectly denied access and $N$ is total number of genuine fingerprint tested.

$$ FAR = \frac{I}{N} $$

(12)

where $I$ the number of imposters’ fingerprints is incorrectly granted and $N$ is the total number of genuine fingerprint tested.

$$ SA = \frac{M}{P} \times 100 $$

(13)

where $SA$ is the system accuracy, $M$ is the total number of organized fingerprint image samples and $P$ is the total number of fingerprint samples.

$$ x = \left\{ {\begin{array}{*{20}c} {x_{ij}^{0} } & {K\left\{ {\left| {x - y} \right| \le D_{1} } \right\} \ge N_{1}^{th} } \\ {x_{ij}^{n} } & {else} \\ \end{array} } \right. $$

(14)

$D_{1}$ is adopted as the maximum depth difference between the similar $x$ and $y$ pixels and is often assumed to be eight pixels in the neighborhood. $N_{1}^{th}$ is one, as every pixel is assumed to be similar to at least one pixel, and K is the number of $y$ pixels that satisfies Eq. (14) while the distorted pixel is eliminated.