Personal Authentication based on Angular Geometric Analysis using

Proc. of Int. Conf. on Advances in Communication, Network, and Computing
Personal Authentication based on Angular Geometric
Analysis using Finger Back Knuckle Surface
K.Usha1 and M.Ezhilarasan2
1
Research Scholar, Dept., of CSE,
Pondicherry Engineering College, Pondicherry, India.
2
Associate Professor, Department of Information Technology,
2
Pondicherry Engineering College, Pondicherry, India.
1
[email protected],[email protected]
1
Abstract— In this paper a new authentication system using Finger Back Knuckle Surface is
examined. This paper is intended to propose an effectual method for personal
authentication using Finger Back Knuckle Surface. This method take advantage of the less
number of features extracted from the Finger Back Knuckle Surface and the Angular
geometric analysis is done to mine the distinguishable feature information which is
computationally less complex. This is achieved by the determination of convex curves from
the finger back knuckle surface. From the recognized feature curves, the subset features like
Knuckle Edge points, Knuckle Tip points and Knuckle Base Points are identified. From
these features, geometrical structures like scalene triangles were constructed to obtain
feature information in terms of angle. This method reduces critical problems that arise due
to the extraction of more number of features. Also reduces the computational complexity of
the feature extraction and recognition process.
Index Terms— Finger Back Knuckle Surface, Napier’s rule, Correlation Coefficient,
Matching Score Level Fusion.
I. INTRODUCTION
The main idea behind the biometric systems is to identify the persons using pattern recognition algorithms on
one or more physiological and behavioral characteristics which can be called as biometric traits [1]. This
biometric system can function in two different modes, Verification Mode and Identification Mode. In
Verification mode, an individual claims an identity by presenting the biometric trait to the sensor and for
recognition one to one matching with the stored pattern template is done. In identification mode, an
individual is authenticated by presenting the biometric trait to the sensor and by comparing the obtained
pattern template with all registered template of the system, this form the one to many matching.
Hand based biometric recently has drawn a prevalent attention of researchers, since it is advantageous in
terms of high user acceptability and more accuracy with low level of processing. Because of this, numbers of
new modals are developed based on hand based biometrics. One such modal is Finger Back Knuckle Surface,
can also be known as finger knuckle print. Woodard and Flynn 2005 [2] have introduced Finger knuckle print
as a biometric trait by capturing it in a 3D sensor. Feature extraction for identification is done by extracting
the curvature shape information of the finger knuckle print. The limitation of this paper is that 3D data
© Elsevier, 2013
processing is computationally expensive. This limitation is overcome [3] by capturing a finger knuckle print
using 2D acquisition device and uses line orientation coding scheme for feature extraction. Following them
many researchers were done biometric authentication system using finger knuckle print based on location and
line features [4],[5].
This paper we recommend a new approach for personal authentication based on Angular Geometric Analysis
Method which is implemented on Finger Back Knuckle Surface. This method take advantage of the less
number of features extracted from the Finger Back Knuckle Surface which is computationally less complex.
The paper is organized in the following manner. This paper commences with the brief survey of the existing
methods of the Finger Knuckle Print based authentication in Section 2. The details of the implementation
model of personal authentication system using proposed angular geometric technique for feature extraction
are explored in Section 3. The Section 4 describes the Correlation Coefficient technique used for
classification and Section 5 describes the Match score level fusion scheme to take the final decision for
authentication. Section 6 demonstratesthe results of the Verification and Identification experiments
conducted using the proposed method. This paper concludes with the summary of the results and the future
scope for the work in Section 7.
II. EXISTING WORK
A number of techniques for personal authentication based on hand based biometric traits such as Finger Print,
Palm Print, Finger back Knuckle surface, Palm side Knuckle surface have been proposed in the literature.
Most of the Hand based biometric trait recognition methods primarily employ two types of Recognition
methods such as, Geometrical Analysis, and Texture Analysis. Geometrical based analysis method for feature
extraction schemes are different edge detection methods are used to extract lines, wrinkles, and ridges in the
Finger Knuckle surface. The extracted edges directly or being represented other formats. Texture based
extraction schemes, in which images are divided into blocks, where the variations exist in either blocks of
images are extracted. Some of the existing works in the literature on hand based biometrics were discussed
below.
The development of a new biometric system which uses Finger back surface imaging as a biometric trait for
personal authentication is presented in this paper [6]. In this feature information of knuckles were extracted
using both texture analysis and geometrical feature analysis to get better performance in the personal
authentication. The preprocessing is done by binarizing the image to obtain contour pixels in the image.
Along with the contour pixel point, palm wrist point M is identified. Various extreme finger base point were
also identified the geometrical information from the finger back knuckle surface imaging. Six geometrical
features such as Finger Length, Three different Finger widths, Finger Perimeter and finger area were
computed from each finger. From the acquired knuckle image totally 24 geometric features were obtained.
Image is normalized using Min Max normalization scheme and Z-score normalization scheme to obtain
Texture information by enrolling the discrepancy in the band of knuckle. This method of texture analysis can
be termed as appearance based approach. This is implemented by means of Principal Component Analysis
(PCA), Independent Component Analysis (IDA), and Linear Discriminate Analysis. Score generation is done
by calculating Euclidean distance between the obtained coefficients from reference and input samples. Fixed
fusion rules such as SUM and PROD were used to merge obtained matching scores. Experiments were
conducted with newly created database with 105 users 630 images implementing all the proposed methods.
A new approach to personal authentication system by utilizing the hand vein structures and knuckle shape
information is investigated in [7]. The Hand back surface is captured using a low cost, contact less Infra red
imaging. From the captured hand back imaging, the structure of the vein is studied using key point
triangulation method. The acquired image is subjected to adaptive histogram equalization for enhancement of
the image. The extraction of key points was identified in the vein structure of the binarized finger back
surface image. From the identified key points the feature extraction approach is implemented using Delaunay
triangulation. This paper also focuses on incorporating the simultaneously extracted of knuckle shape
information to achieve better performance.
A new personal authentication system using finger Knuckle print based on the quality of the trait is
contributed in [8]. This paper focuses quality based analysis for the hand biometrics. The basic idea behind
this work is the quantification of quality of the data acquired from the user. This quantified information is
taken to estimate consistent matching score. The single hand image is taken as input trait to the biometric
system. From the acquired hand image, simultaneous extraction of features like, palm print texture and hand
shape information. Experiments were conducted using PolyU database for palm print to show the EER value
52
has been decreased when there is a fusion of palm print and hand information using quality dependent fusion.
Another set of experiments were conducted by incorporating the Finger Knuckle print features to obtain high
accuracy.
A new approach for biometric identification system based on Finger Knuckle print recognition method is
explored in [9].The acquired image has been processed to extract two different feature vectors, one consisting
of gray scale image and other one consisting of Gabor image of the gray scale intensity. The extracted ROI
image from the acquired image is divided into 22 parts in such a way that each part consists of 1100 pixels. A
statistical based method known as Average Absolute Deviation is used to quantify the gray scale information
in the ROI image. This AAD method quantifies the local information of the sub image. The most prominent
feature was selected by applying the combination of Principal Component Analysis and Linear Discriminate
analysis in the information obtained from the sub image. Matching between the reference Image and Input
Image is done based on the Euclidean distance.
An efficient multimodal biometric system by fusing palm print and knuckle print at matching score level to
provide a high security feature [10]. This is achieved by means of efficient matching algorithm known as
Phase – correlation function. Linear phase shift in the frequency domain of palm print images are given by
means of Discrete Fourier Transform. Fourier transform between input image and reference image to define
the phase shift between them. The identified phase shift are taken inverse discrete Fourier transform to define
the cross correlation between phase components. This type of matching process using correlation between
phase components can be called as Phase correlation function. The observations based on Phase correlation
functions is done by calculating cross correlation and locating similarity between the two images reference
image and input image . The same method is used to extract similarity measure between knuckle print
images.
From the study conducted with the existing works in the Literature for Recognition methods of Finger
Knuckle Print, the analysis shows some of the limitations. In some of the texture analysis methods the
differences in appearance were recorded to generate unique information which will lead to degrade in the
performance. The accuracy of the system is mainly depends upon the knuckle segmentation methodology.
The more accuracy can be obtained with the more accurate segmentation, which gives rise to two different
tradeoffs – User acceptability and Computational Complexity. According to the existing works with the low
quality images the accuracy can be obtained by more number of partitioned blocks which in turn increases
computational complexity. It has been identified that there is a need of the multimodal biometrics approach
which can be extended by explicitly considering the sample of the input biometric signals and weighting the
various pieces of evidence based on objective measures of multiple features of the biometric traits. The
measurement of feature information can be done by means of geometric analysis which is computationally
economic.
III. THE IMPLEMENTATION MODEL
This paper innovates a method for Personal authentication using Finger Back Knuckle surface (FBKS) as
biometric identifier as shown in the fig.1. The proposed system gets the bending surface of FBKS using
contact less imaging. The images of Right index finger, Left index finger, Right Middle finger and Left
Middle finger are given as input to the proposed model. As a first step, Preprocessing of the FBKS image is
done to extract region of interest. The extraction of feature information from the FBKS is done by means of
Angular Geometric Analysis Method. The extracted feature information of the registered finger is recorded
in the form of vector. This vector information is passed to the matching module which performs matching
between the reference and input vector using correlation coefficient. There generated matching scores for
different fingers were fused for matching score level fusion to take the final decision of authentication.
A. Locating Contours in Finger back Knuckle Surface (FBKS)
The establishment of contours is done by subjecting the FBKS to the Canny EdgeDetection algorithm [1112].This algorithm identifies the curved feature of FBKS and produces the output in theform of convex
curves as shown in the following figures fig.2(a) and fig.2(b). From the identified convex curves the knuckle
edge points were calculated were calculated using Knuckle Edge Calculation Algorithm.
Knuckle Edge Calculation Algorithm
i) Establish sequence of points that the edge convex curve passes through.
ii) Calculate the intensity level of various points [1…Pn] in the established line.
iii) Restrain the pixel Pi which is not maximum where i=1, 2, 3…n.
53
iv) Call edge threshold method to identify the edge pixel which is above the threshold intensity level
v) Identified edge pixel of the right hand side curve is named as a REP (Right Edge Point).
vi) Identified edge pixel of the left hand side curve is named as a LEP (Left Edge Point).
Fig.1: The Block Diagram of Implementation Model for Personal Authentication using FBKS
(a)
(b)
Figure 2: (a) Image obtained by applying Canny Edge Detection Algorithm in FBKS Image (b) Convex curves of FBKS
The other pixels points were identified from the shape boundary information of the FBKS. The Base points
such as BP1, BP2, BP3, and BP4 of the left side curve are stored in a separate vector and similarly for the
based points of right side curve BP5, BP6, BP7 and BP8.
B. Angular Geometric Analysis Method (AGAM)
The proposed Angular Geometric Analysis Method (AGAM) on FBKS efficiently extracts feature
information from the identified knuckle contours. This method constructs the geometrical structures like
scalene triangles to obtain angular information from the recognized knuckle quality points.
Steps involved in Angular Geometric Analysis Method
(i) Construct a straight line joining BP1 and BP2 and another straight line joining BP3 and BP4 in the left side
of FBKS as shown in the fig.3. The Magnitude of the line (i.e) the distance between the respective base
points were calculated and store in the Reference vector Vrefwhich is of Nx3 dimension.
(ii) Construct a scalene triangular structure by joining the set of points such as BP3, LEP and BP4. Construct
similar triangular structure for right side curve by joining the set of points BP5, REP and BP6. The magnitude
of the line joining the points is calculated by measuring the distance between the points represented as a, b
and c as shown in the fig 4.
Fig.3: Construction Geometrical Structures in the identified contours of FBKS
The angular information of the triangle is calculated by means of Napier’s rule in the following fig.4
54
Fig.4: Representation of Triangular feature
iii) As per the Napier rule [13] for scalene triangle, it has been proved that in (1), (2)
a  b  c  2s
(1)
s=
abc
;
2
(2)
Thus angle C can be computed as follows
 C =2*sin-1
(s - a)(s - b
a
(3)
iv) This computed angle for a Left Index Finger can be named as Primary Knuckle angle for Left Index
Finger (PLIF) and Secondary Knuckle angle Left Index Finger (PSIF). Similar angle calculated for all other
fingers can be given as Primary Knuckle angle for Left Middle Finger (PLMF), Secondary Knuckle angle for
Left Middle Finger (SLMF), Primary Knuckle angle Right Index Finger (PRIF), Secondary Knuckle angle
Right Index Finger (SRIF) Primary Knuckle angle for Right Middle Finger (PKIF), Secondary Knuckle angle
for Right Middle Finger (SKIF). These obtained angles were converted in to radians for simple classification
and stored in the Reference Vector (Vref).
IV. CLASSIFICATION SCHEME
Classification scheme is defined and developed using correlation coefficient [14] computed between
reference and Input image. Correlation coefficient based classification gives the strength of similarity
between the values of reference vector values Vref and the values of Vector obtained from the Input vector
VInp.
The population correlation (ρ) between the values from the two different vectors Vref and Vinp can be defined
as

 v refi  v ref
ρ

v refi  v ref 
v inpi  v inp 
2
 i ( V inpi  V inp)
(4)
2
Where, ρ is called the Product Moment Correlation Coefficient or simply the Correlation Coefficient. It is
a number that summarizes the direction and closeness of linear relations between two values. The sample
value is called r, and the population value is called ρ (rho). The correlation coefficient can take values
between -1 through 0 to +1. The sign (+ or -) of the correlation defines the direction of the relationship.
When ρ=0, r is distributed around 0 symmetrically, and mean of the sampling distribution does equals the
parameter. As ρ approaches +1 or -1, the sampling variance decreases. If the value approaches to 1, the
shaping variable approaches to Zero.
From this analysis, it can be stated that
i) When the ρ value is negative, indicates there no similarity relationship between the values of the
two different vectors, hence the corresponding Input is said to be rejected.
ii) The positive values which lie in the intervals of 0 to 0.5 also indicated weak relationship between
the values; hence the corresponding input can also be rejected.
iii) The positive values which lie in the intervals of 0.5 to 0.83 also indicate somewhat stronger
relationship between the values; hence the corresponding input can be accepted with the maximum
threshold of 0.28.
iv) The positive values which are greater than 0.83 indicate strongest similarity relationship between the
values; hence the corresponding input can be directly accepted.
55
V. FUSION SCHEME
Matching Score level fusion [15] scheme is adopted to consolidate the matching scores produced by Knuckle
surface of the single finger. In the Matching score level, different rules can be used to combine scores
obtained by biometric systems. All these approaches provide significant performance improvement. In this
paper, weighted rule has been used. In the weighted rule, for example if S1,S2,S3 and S4 represent normalized
score obtained from Finger back knuckle surface of Left Index Finger, Right Index finger, Left middle finger
and Right middle finger respectively. The final score SF is computed using (5)
(5)
Sp  W1 S1  W 2 S2  W 3 S3  W 4 S 4
where
,
,
and
are weights associated with the units define in the (6)
wi 
(6)
EER i
 ik  1 EER k
is Equal Error Rate obtained by considering single feature extraction method.
VI. EXPERIMENTAL RESULTS AND DISCUSSION
To evaluate the performance of proposed bimodal biometric scheme, publically available database containing
Finger Knuckle prints have been used. The public PolyU FKP database [16] is used in the paper for testing.
The FKP database consists of images with size 220×110, captured from 165 volunteers. In this multiple
feature information are exploited and the results are given in the form of comparison and combination using
the same database. The experiments were conducted and the results were analyzed with performance
parameters Genuine Acceptance Rate andFalse Rejection Rate. In this method more accuracy is achieved in
an efficient way.
Experiment 1: Verification Mode
The goal of the experiment is to evaluate the performance of the Personal authentication system in the
verification mode. Verification mode aims to identify the person is the one he/she claims to be. First,
Performance is evaluated for single finger type, then fusion of two different finger type and all the four finger
types. In this experiment, we compare our proposed method of feature extraction with three other existing
methods.
Finger Geometric Method: In this method, six geometrical feature information is obtained from the Finger
Knuckle Print. Along with that Knuckle texture information identified by means of PCA, ICA and LDA
algorithms. The result obtained for the fixed dimension of feature vector.
Palm print and Hand Shape Method: In this method, Palm print texture information is obtained by applying
PCA algorithm to the entire image. The Hand shape information is obtained by various control points in the
acquired hand image. In addition to that, Finger knuckle information is used to improve the performance.
Hand Vein Geometric Method: In this method, Delaunay triangulation is used to extract hand vein structure
information and geometrical methods for extracting the Knuckle shape information. Using this method, this
limits to selected set of feature vector. The following table I illustrates the comparison of AGAM with the
existing work.
Experiment 2: Identification Mode
The goal of the experiment is to evaluate the performance of the Personal authentication system in the
Identification mode. Identification mode aims to determine the identity of the person who may or may not
present in the database. First, Performance is evaluated for single finger type, then fusion of two different
finger type and all the four finger types. In this experiment, we compare our proposed method of feature
extraction with three other existing methods.
Palm Print and Knuckle Print (PKP): In this method, PCF is used to find the correlation between input and
reference images. The modalities are subjected to DFT to identify he phase difference. The result obtained
for the fixed dimension of feature vector.
Finger Knuckle Print (FKP): In this method, the quantification of the gray scale information is done by
means of AAD method. The combination of PCA and ICA were used to utilize the most prominent
information the FKP.
The following fig. 6 depicts that AGAM outperforms the other three methodologies namely PKP and AAD in
terms of Genuine Acceptance Rate.
56
TABLE I: ILLUSTRATION OF COMPARISON RESULTS OF ACAM WITH E XISTING WORK
The following fig 5 depicts that AGAM outperforms the other three methodologies namely FGM,PHM and
HVG in terms of Genuine Acceptance Rate.
100
GenuineAcceptanceRatio(GAR)%
98
96
94
92
FINGER GEOMETRIC METHOD
PAlM PRINT AND HAND SHAPE METHOD
HANDVEIN GEOMETRIC METHOD
FBKS ANGULAR GEOMETRIC METHOD
90
88
0
10
20
30
40
50
60
70
Fals e Ac ceptance Ratio(FAR)%
80
90
100
Fig 5: The Comparison between FAR and GAR for Experiment 1
100
GenuineAcceptanceRatio(GAR)%
99
98
97
96
95
94
93
FINGER KNUCKLE PRINT
PALMPRINT & KNUCKLE PRINT
92
FBKS ANGULAR GEOMETRIC METHOD
91
0
10
20
30
40
50
60
70
False Acc eptanc e Ratio(FAR)%
80
90
100
Fig 6: The Comparison between FAR and GAR for Experiment 2
The proposed technique for personal recognition using Finger back knuckle surface achieves better
performance than the existing techniques in the literature.
The discussions about the results of an experimental study were presented below.
(i) Graphical illustrations suggest that there is significant improvement in performance using the
proposed method.
57
(ii) In the conducted experiments, there is no single fusion combination best performing type. Different
fusion combination performs better during the different experimental types.
(iii) From this experiment, It is also been proved that the Finger back knuckle surface as a biometric
identifier achieves better performance similar to that of other biometric traits which has been under
research for longer time such as finger prints.
(iv) When using the finger knuckle prints and finger prints along with palm prints in the Verification
mode experiment there was 1.5% difference in performance.
(v) Also using the finger knuckle print and finger prints fused with palm prints in the Identification
mode experiment, there was 1.3% difference in the performance obtained.
VII. CONCLUSION
This paper has investigated a new approach to achieve better performance in the hand based biometric system
using finger back knuckle surface. This paper innovates a method known as Angular geometric analysis
Method for feature extraction by incorporating less number of features in the Finger Back knuckle surface.
The reliable matching scores were developed using matching score level fusion of four different finger back
knuckle surfaces. Experimental results prove that the proposed method produces better results with the
reduced computational complexity.
REFERENCES
[1] Bolle R. M.,Connell J. H,Pankanti S., Ratha N. K., Senior A.W.:Guide to Biometrics.New York: SpringerVerlag,(2003).
[2] Woodard. D.L and Flynn P.J.: Finger surface as a biometric identifier.Computer Vision and Image
Understanding,Vol.100(3), pp.357-384.(2005).
[3] Kumar.A and Ravikanth.C.: Personal authentication using finger knuckle surface. IEEE Trans. Information
Forensics and Security,Vol.4 (1), pp.98-109.(2009).
[4] Kumar.A and Zhou.Y.: Personal identification using finger knuckle orientation features.Electronic Letters,Vol.
45(20), pp.1023-1025. (2009).
[5] Kumar.A and Zhou.Y.:Human Identification using Knuckle Codes. In Proceedings of BTAS'09.(2009).
[6] Ajay Kumarand Ravikanth.C.: Personal Authentication Using Finger Knuckle Surface. IEEE Transactions on
Information Forensics and Security,Vol. 4,No.1(2009).
[7] Ajay Kumar and K.Venkata prathyusha.:Personal authentication using Hand vein Triangulation and Knuckle shape.
IEEE Transactions on Image processing, VOL 18, No.9,(2009).
[8] Ajay Kumar and David Zhang.:Improving Biometric Authentication Performance from the User Quality. IEEE
Transactions on Instrumentation and Measurement. Vol.59, No.3 March (2010).
[9] Shariatmadar.Z.S.; Faez.K.: An efficient method for Finger-Knuckle-Print recognition based on information fusion.
IEEE International Conference on Signal and Image Processing Applications (ICSIPA), pp: 210 – 215 (2011).
[10] Saigaa, M,Meraoumia, A,Chitroub, S,Bouridane.A.: Efficient person recognition by Finger-Knuckle-Print based on
2D Discrete Cosine Transform. ICITeS,1 – 6,(2012).
[11] Paul Bao, Lei Zhang, and Xiaolin Wu.: Canny Edge Detection Enhancement by Scale Multiplication.: IEEE
Transactions On Pattern Analysis And Machine Intelligence, Vol. 27, no. 9,(2005).
[12] Zhang.L, Zhang.L, Zhang.D, and Zhu.H.: Online finger knuckle-print verification for personal authentication.
Pattern Recognition,Vol. 43, pp.2560-2571,(2010).
[13] Daniel Zwillinger.: Standard Mathematical Tables And Formulae. Chapman & Hall/CRC A CRC Press
Company(2003).
[14] David Shen, Zaizai Lu.: Computation of Correlation Coefficient and Its Confidence Interval in SAS. SUGI 31, Paper
170-31,(2005).
[15] Madasu Hanmandlu1,Jyotsana Grover1,Vamsi Krishna Madasu2,Shantaram Vasirkala.: Score Level Fusion Of
Hand Based Biometrics Using T-Norms.IEEE (2010).
[16] Polyufkpdatabase, http://www.comp.polyu.edu.hk/˜biometrics/FKP.htm.
58