See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/320044516

Real-Time Multimodal Biometric User Authentication for Web Application

Access in Wireless LAN

Article  in  Journal of Computer Science · September 2017

DOI: 10.3844/jcssp.2017.680.693

CITATIONS READS

2 396

3 authors:

Sanjay Kumar Surjit Paul

National Institute of Technology, Jamshedpur Indian Institute of Technology Kharagpur
22 PUBLICATIONS   32 CITATIONS    13 PUBLICATIONS   9 CITATIONS   

SEE PROFILE SEE PROFILE

Dilip Kumar Shaw

National Institute of Technology, Jamshedpur
12 PUBLICATIONS   13 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Mangement Development Program Management System View project

INDUSTRIAL RELATIONSHIP MANAGEMENT View project

All content following this page was uploaded by Sanjay Kumar on 24 December 2017.

The user has requested enhancement of the downloaded file.

 Journal of Computer Science

Original Research Paper

Real-Time Multimodal Biometric User Authentication for

Web Application Access in Wireless LAN
1
Sanjay Kumar, 1Surjit Paul and 2Dilip Kumar Shaw
1
Department of Computer Science and Engineering, NIT Jamshedpur, Jharkhand, India
2
Department of Computer Application, NIT Jamshedpur, Jharkhand, India

Article history Abstract: Web applications store trustworthy information and can be
Received: 29-07-2017 accessed online through wired or wireless network. Authentication is one of
Revised: 01-08-2017 the major challenges to access these web applications. With varying level
Accepted: 25-09-2017 of sensitive data stored in web application, the concept of level of
Corresponding Author:
authentication can be introduced using biometric traits. This paper proposes
Sanjay Kumar biometric-based multi-modal authentication system with four levels of
Department of Computer securities. Level 1 uses user name and password only; Level 2 uses
Science and Engineering, NIT fingerprint with user name and password; Level 3 uses fingerprint and face
Jamshedpur, Jharkhand, India with user name and password; Level 4 uses fingerprint, face and iris with
Email: [email protected] user name and password.

Keywords: Trustworthy, Authentication, Biometric, Security, Multimodal

Introduction traits, they cannot be stolen, guessed, forgotten or easily

forged. To provide better security to the web
Due to advancement of web technology, it is possible applications, biometrics traits can be encrypted further.
to design and develop web application that can fulfill the The unimodal biometric authentication system uses
customer demand 24×7. These web applications are used only one physiological or behavioral trait for
to store the commercial, personal, military and authentication along with username and password. Each
information related to government etc. and can be access unimodal biometric authentication system has certain
online. To access this information in real time a proper limitations due to following reasons: Noise involved in
verification or authentication method is required. The sensor data, lack of universality and individuality in
various approaches used for authentication are as follows: selection for user authentication. To overcome the
limitations of unimodal biometric authentication system,
• Possession: The use of identity card multimodal biometric authentication system can be
• Knowledge: The use of PIN and password devised to enhance the level of security by incorporating
• Inherent: Biometric traits like physiological multiple biometric traits.
or behavioral The proposed multimodal system uses biometric
traits like fingerprint, face, iris, along with user id and
The first two approaches are widely used. However password. Using these biometrics traits, four levels of
the tokens such as ID cards can be stolen lost or authentication are introduced. To access web application
duplicated. For the second approach, one needs to containing less sensitive information Level 1
remember lengthy and complex passwords which area authentication is used whereas level 4 is used for web
difficult task. Further, in these two approaches, a person application containing most sensitive information. Level
identity is verified or authenticated based on what he/she 1 uses user id and password; Level 2 uses fingerprint
knows or possesses rather than who he/she is. Due to with user id and password; Level 3 uses fingerprint and
these reasons the first two approaches are susceptible to face with user id and password; Level 4 uses
fraudulent attack. Biometric, an emerging technology fingerprint, face and iris with user id and password.
can be used to establish identification or verification of To increase the security, biometric traits is further
users through two types of inherent biometric traits like encrypted using AES algorithm.
physiological traits (finger- print, face, iris etc.) or Rest of the paper is organized as follows: Section 2
behavioral traits (voice, gait, signature etc.). It also Background work is introduced, Section 3 introduced the
requires the person to be present during time of proposed work; Section 4 introduced the
authentication. Also, due to uniqueness of the biometric implementation of the proposed work, Section 5 deals

© 2017 Sanjay Kumar, Surjit Paul and Dilip Kumar Shaw. This open access article is distributed under a Creative Commons
Attribution (CC-BY) 3.0 license.
Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

with Experimental Evaluation and section 6 deals with simultaneously in order to determine user’s identity.
conclusion and future work. Biometric data is irrevocable (Zhang et al., 2009). Some
algorithms for revocable biometrics exist, but they
Related Work seriously affect verification precision. Therefore, there is
a need for extra layer of security, because if an attacker
Recently the Wireless LAN (WLAN) compromises raw biometric data, it cannot be replaced.
communications has become one of the fastest growing The identification and matching process takes less than
sectors in telecommunication and network industry. The one second to complete (Jain et al., 1998). This depends
wireless network provides many advantages over the on the environment where the system is hosted as there are
wired networks. However, the management of such many factors that delay the execution of the program such
wireless network proves to be challenging. Due to as bridge in network transmission. The system determines
increase in the web application attacks, highly reliable the users identity by comparing the match score to a
and convenient personal identification and verification threshold set by the administrator (Jain et al., 2005;
technology are vital in our society today (Fry and Ratha et al., 2007; 2001). To secure access of web pages
Dunphy, 2009). The ancient Babylonian conducted a prototype of biometric based authentication system
business transaction by pressing the tips of their based on hand geometry is used (Jain et al., 1998). The
fingertips into clay (Lee and Gaensslen, 2001). The use security issues, a set of security automated tools and
of fingerprint as a valid means of identification was methodology are discussed in each stage of SDLC
formally accepted by the law-enforcement agencies in (Teodoro and Serrao, 2011). Fingerprint based student
the early 20th century (Šošević et al., 2013). Even two monitoring system is developed using Java technology
samples of biometric data gathered from the same person and MySQL to evaluate the performance and evaluation
are never the same, due to sensor noise, aging and of biometric based web application (Okafor and
imperfect acquisition conditions (Jain et al., 2004). Ogbuabor, 2013). A real time multimodal biometric
Therefore, there is always a possibility of biometric authentication system using java for secure access to
system error. There are two types of recognition errors in internet banking web page based on password and
fingerprint biometrics: False accept rate and false reject fingerprint (Cătălin et al., 2015).
rate. Fingerprint identification system performance is
measured in terms of its False Accept Rate (FAR) and Proposed Work
False Reject Rate (FRR). If a non-matching pair of
fingerprints is accepted as a match, it is called a false Security is a one of the major concern for all types of
accept while if a matching pair of fingerprints is rejected application access stored in a server either in wired or
by the system, it is called a false reject (NIST, 2000; wireless network. Single level security is not sufficient
Merati, 2011). It has been reported (Jain et al., 1998) to in today’s scenario as it can be easily breakable due to
the U.S. Congress that approximately two percent of the advancement of parallel computing systems. The level of
population does not have a legible fingerprint and security can be enhanced by adding different biometric
therefore cannot be enrolled into a fingerprint biometrics traits in authentication level for web applications stored
system. Multimodal biometrics is a possible solution for in web server. In the proposed work we have devised
improving biometric system precision (Jain et al., 2008). multimodal biometric authentication system architecture
In multimodal biometrics, different modalities are taken for web application as shown in Fig. 1.

Fig. 1. Proposed system architecture

681
Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

In the proposed architecture, web applications haarcascaded classifier technique is used and for iris
hosted in the server or servers in a wireless LAN can template: Hough transformation is used for iris
be accessed through wireless client after its localization and Doughman Rubbersheet algorithm is
authentication by Advance Authentication Server used for iris normalization. Figure 3 shows the
(AAS). The proposed scheme consists of two phases preprocessing and feature extraction steps used by AAS.
i.e., enrollment and authentication.
Authentication Phase
Enrollment Phase
During authentication phase, a data sample of the
In this phase, an enrollment center is used at the user claiming an identity is acquired after passing
client end to enroll the user. During enrollment, raw through appropriate sensors. This data sample is referred
biometric data of the user is acquired using the sensor. to as query sample. These query samples are passed
To capture fingerprint, Mantra MFS100, a high quality through preprocessing and feature extraction module to
USB electronic fingerprint sensor is used; to capture acquire the salient features. These salient extracted
face, Logitech Camera is used; and for iris 3M Cogent features must be same with those extracted features at
scanner is used. The scanned version of biometric image the time of enrollment. The features of the query sample
is sent to AAS for further preprocessing and feature are compared with the decrypted information of a
extraction to get standard template from raw image. The claimed identity. The process of comparison of query
standard templates is further encrypted through AES sample with the decrypted template stored at the time of
encryption technique and stored in database along with enrollment and the output score of the process is referred
client’s unique id. Figure 2 shows the enrollment steps. to as matching and matching score respectively. The type
of the matching score is either similarity or distance
Preprocessing and Feature Extraction expressing the similarity or dissimilarity of the query
In the preprocessing step, raw biometric template of sample to the stored template. The decision of accepting
fingerprint, face and iris is passed through the or rejecting the identity claim is made by comparing the
Median/Gaussian filter to remove the noise and to matching score with the threshold. For fingerprint
enhance the quality of the image. For generating matching the optimized threshold calculated as 14000;
thumbprint template: Minutia based feature extraction for face the optimized threshold is calculated as 10; and
technique is used, for face template: OPENCV for iris optimized threshold is calculated as 30.

Fig. 2. Advance Authentication Server (AAS)

682
Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

Fig. 3. Preprocessing and feature extraction technique

Advance Authentication Server does enrollment matching is used. If both are matched successfully then
and authentication process. It grants access to only user is authenticated by the AAS and corresponding
authenticated users. web application is loaded in the browser.
The proposed model uses four level of authentication.
During different levels user authentication are performed Level-3 Authentication
as given below: To access web application using Level-3 requires
Level-1 Authentication matching of thumbprint, face along with user id and
password one by one by AAS. For each case, thump print,
AAS matches the username and password of the
claimed identity with hashed password generated using face along with user id and password of the claimed identity
password along with salt stored in the database. Figure 4 is matched with the decrypted stored claim identity during
shows the authentication using user name and password. enrollment. For face matching, PCA based face recognition
algorithm is used. If all three are matched successfully then
Level-2 Authentication only user is authenticated by the AAS and corresponding
To access web application using Level-2 requires web application is loaded in the browser.
matching of thumbprint along with user id and password Level-4 Authentication
one by one by AAS. For each case, thump print along
with user id and password of the claimed identity is To access web application using Level-4, requires
matched with the decrypted stored claim identity during matching of thumbprint, face, iris along with user id and
enrollment. For thumb print, minutia based template password one by one by AAS. Here, thump print, face,

683
Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

iris along with username and password of the claimed successfully then only user is authenticated by the
identity is matched with the decrypted stored claim AAS and corresponding web application is loaded in
identity during enrollment. For iris, PSNR based iris the browser. Figure 5 shows the level wise
template matching is used. If all four are matched authentication process.

Fig. 4. Authentication using hashed password using salt

Fig. 5. Enrollment and level wise authentication

684
Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

Implementation Screenshots of the Implemented System

This section deals with the technologies used and The different screen shots for enrollment and
screenshots of the implemented system. authentication of users for different levels of user
authentication are shown in below figures from Fig. 6-4.
Technologies Used
Level-1 Authentication
Java Technology
In Level-1 authentication, user has to enter only their
It is used to develop client and server side interface. username and password credentials into the
Remote method invocation is used to access different authentication interface. If their credentials do not match
distributed methods present in the server interface. then it shows “invalid credential” message otherwise it
will show the message “you are authenticated” and
JSP & Servlet Technology redirect to corresponding web application as depicted in
It is used to develop web application interface to the below figures.
access the web applications present in the server. Web Level-2 and Level-3 authentication methods uses
applications require different level of authentication. thumb and face biometric traits for authentication. In
Level-4 authentication we are using three physiological
MySQL Database biometric traits i.e., thumb, face and iris for user
Using MySQL database, user enrollment details are authentication along with username and password. The
stored, which is used during the authentication phase. screen shots for Level-4 are as shown below.

Fig. 6. Screenshot of the authentication server interface

Fig. 7. Screenshot of user enrollment interface

685
Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

Fig. 8. Screenshot of Level 1 authentication

Fig. 9. Screenshot of invalid user authentication

686
Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

Fig. 10. Screenshot of successful authentication in Level 1

Fig. 11. Screenshot of redirected web application

Fig. 12. Screenshot of successful authentication in level 4

687
Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

Fig. 13. Screenshot of redirected web application

Fig. 14. Screenshot of authentication time in Level 4

Fig. 15. Encryption time of biometric template during enrollment

688
Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

Fig. 16. Computation time Vs level of authentication

Experimental Evaluation and Result mathematical based performance evaluation and

graphical based performance evaluation.
Time required to access web applications hosted in
the web servers using four level of authentication can be Mathematical Based Performance
expressed using following equations: The authentication of the biometric system can be
Level 1:TAT 1 = TMHPi (1.1) represented by a decision function and can be expressed as:
accept if s ( x ) > ∆
Level 2 :TAT 2 = TAT 1 + TDTthumb + TMT 2 (1.2) D( x)=   (1.5)
 reject otherwise 
Level 3 :TAT 3 = TAT 2 + TDTface + TMT 3 (1.3) where, ∆ is the threshold score value and s(x) is the
matching score obtained by the comparison of template
Level 4 :TAT 4 = TAT 3 + TDTiris + TMT 4 (1.4)
generated during authentication phase with the decrypted
Where: template of the encrypted template stored in the database
TAT1, TAT2, = The Total authentication time during enrollment.
TAT3 and TAT4 for level1, level2, level3 and During decision making process, two most
level4 respectively common types of error may occur are: False
TMHPi = Time to match the hashed Acceptance (FA) Error and False Rejection (FR)
password Error. FA Error may occurs when system falsely
accept the imposter (A person claiming an identity
TDTthumb, TDTface = Template decryption time for
other than their own) whereas FR Error may occurs
and TDTiris thumb, face and iris respectively
when the system falsely rejects the claimant’s
TMT2, TMT3 and = Template matching time for
(Genuine user) during authentication. The normalized
TMT4 thumb, face and iris
version of FA and FR are known as FAR and FRR
Authentication time (in milliseconds) for four level of respectively. They are often used in the performance
authentication is shown in Table 1 and 2 shows the analysis and are defined as:
Encryption time and their corresponding graph are
FA ( ∆ )
shown in the Fig. 15 and 16 respectively. FAR ( ∆ ) = (1.6)
During enrollment 100 users were enrolled, whose Ni
biometric traits like finger print, face and iris were captured FR ( ∆ )
and stored in the MySQL database. The screenshot of the FRR ( ∆ ) = (1.7)
Nc
records of sixteen users are shown in Fig. 17.
where, FA and FR give total no of counts for false
Performance of the Purposed System
accepted access and false rejected access respectively;
The system performance of the proposed biometric and Ni and Nc represent total number of imposter and
based multimodal system can be done using client access.

689
Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

FAR and FRR are the functions of threshold ∆ and GAR ( ∆ ) = P ( S ≥ ∆ | Client ) =1 − FRR ( ∆ ) (1.12)
can be expressed in terms of class distribution of
matching score. Let fC(s) = Pr(S=s|Client) and fI(s) = Graphical Based Performance Evaluation
Pr(S=s|Imposter) be the probability density function of
client and imposter scores respectively. The FAR and the In order to set the threshold value for fingerprint, face
FRR of the biometric system may be expressed as and iris scanner at which the FAR and FRR are found to
equation 1.8 and 1.9: be minimum, two types of curve known as Receiver
Operating Characteristics (ROC) Curve and Detection
∞ Error Trade-off (DET) Curve are used. ROC curve
FAR ( ∆ ) = P ( S ≥ ∆ Imposter ) = ∫ f I ( s ) ds (1.8)
∆
shows the relationship between FAR and GAR whereas
∆
DET shows the relationship between FAR and FRR. The
FRR ( ∆ ) = P ( S < ∆ Client ) = ∫ f C ( s ) ds (1.9) ROC curve enables to set threshold values ∆ for
−∞
different biometric traits at the intersection point of
For finding the better result in biometric system, two FAR and FRR curves with respect to different
types of notations True Acceptance (TA) and True threshold values. The DET curve provides the trade-off
Rejection (TR) are used. TA means claims made by clients between the two types error (False Acceptance and
are correctly accepted whereas TR means claims made by False Rejection), which enables the user to select the
imposter are correctly rejected.TAR is also known as GAR. threshold according to the system requirements. For
TAR and TRR can be expressed as follows: testing purpose, we have taken number of imposter (Ni)
= 20 and number of client (Nc) = 100. Figure 18, 20
∞
TAR ( ∆ ) = P ( S ≥ ∆ Imposter ) = ∫ fC ( s ) ds (1.10) and 22 show the ROC curve for fingerprint, face and
∆
iris respectively. Figure 19 shows the threshold value
∆ for fingerprint, which is equal to 14000. Figure 21
TRR ( ∆ ) = P ( S < ∆ Imposter ) = ∫ f I ( s ) ds (1.11)
−∞ shows the threshold value for face, which is greater
than 31 and Fig. 23 shows the threshold value for iris,
The relationship between GAR and FRR can be which is greater than 11. Figure 24-26 shows the DET
expressed as follows: curves of fingerprint, face and iris respectively.
Table 1. Authentication time in various level of authentication using AES 128 bit
Authentication level TMHPi TDTthumb TDTface TDTiris TMT TAT
1 2579 ms NIL NIL NIL NIL 2579 ms
2 2579 ms 10 ms NIL NIL 1749 ms 4338 ms
3 2579 ms 10 ms 1609 ms NIL 4733 ms 8931 ms
4 2579 ms 10 ms 1609 ms 74ms 4902 ms 9174 ms

Table 2. Encryption time of different biometric traits using AES 128 bit technique during enrollment
Encryption Password Thumb Face Left Iris Right Iris Total time
technique hashed time encryption time encryption time Encryption time Encryption time for encryption
AES 128 bit 13 ms 7 ms 862 ms 135 ms 140 ms 1157 ms

Fig. 17. Screenshot of the MySQL database of the enrolled users

690
Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

Fig. 18. ROC for fingerprint Fig. 21. FAR and FRR versus threshold for ace

Fig. 19. FAR and FRR versus threshold for fingerprint Fig. 22. ROC for iris

Fig. 20. ROC for face Fig. 23. FAR and FRR versus threshold for iris

691
Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

Conclusion and Future Work

In our proposed multi-modal biometric authentication
system, concept of multi-level security i.e., Level1, 2, 3
and 4 are employed. In Level1, we have used hashing of
password with salt using Hashed MAC (HMAC) and
SHA, to further strengthen the security of the password.
For the other levels, we have used biometric traits like
thumb print, face and iris. From the graph it is apparent
that as the level of authentication increases the level of
security as well as authentication time also increases.
The ROC enables the user to select a threshold that best
meets system requirement graphically. The DET curve
provide the trade-off between the two types of error
(FAR and FRR) which enables the user to select the
threshold according to system requirement. The lower
the DET curve the better the performance. With varying
Fig. 24. DET curve for fingerprint level of security, people will prefer to do e-transactions,
e-commerce and m-commerce securely.
For future work: Different chaotic encryption
algorithms can be employed for biometric templates
security; the developed model can be extended for securing
websites access through internet; the developed model can
be implemented as mobile application for secure
authentication; the proposed system can be developed as
firmware in router to implement device level security.

Acknowledgement
The authors feel greateful to the anonymous
reviewer for their valuable comments and sugessions
to improve the quality of paper and would like to
thank them from core of the heart.

Author’s Contributions
Sanjay Kumar: Conceptualization, Design and
Analysis Drafting and Critical revision.
Fig. 25. DET curve for face Surjit Paul: Execution, Drafting and Revision.
Dilip Kumar Shaw: Drafting the Manuscript and
Revision.

Ethics
After publication of the paper, if we learn any sort of
errors that changes the interpretation of the research
findings, We are ethically obligated to promptly correct the
errors in a correction, retraction, erratum or by other means.

References
Cătălin, L., V.G. Găitan and V. Lupu, 2015. Security
enhancement of internet banking applications by
using multimodal biometrics. Proceedings of the
IEEE 13th International Symposium on Applied
Machine Intelligence and Informatics, Jan. 22-24,
IEEE Xplore Press, Herlany, Slovakia.
Fig. 26. DET curve for iris DOI: 10.1109/SAMI.2015.7061904

692
 Sanjay Kumar et al. / Journal of Computer Science 2017, 13 (12): 680.693
DOI: 10.3844/jcssp.2017.680.693

Fry, J. and A. Dunphy, 2009. Biometric student NIST, 2000. Summary of NIST standards for biometric
identification: Practical solutions for accountability accuracy, tamper resistance and interoperability.
and security in schools. Report to the United States Congress.
Okafor, F.O. and G Ogbuabor, 2013. Performance and
Jain, A.K., A. Ross and S. Prabhakar, 2004. An
security evaluation of biometric-based web
introduction to biometric recognition. IEEE application. West Afr. J. Industrial Acad. Res.
Trans. Circuits Syst. Video Technol., 14: 4-20. Ratha, N.K., J.H. Connell and R.M. Bolle, 2001.
DOI: 10.1109/TCSVT.2003.818349 Enhancing security and privacy in biometricbased
Jain, A.K., A. Ross and U. Uludag, 2005. Biometric authentication systems. IBM Syst. J., 40: 614-634.
template security: Challenges and solutions. DOI: 10.1147/sj.403.0614
Proceedings of the 13th European Signal Processing Ratha, N.K., S. Chikkerur, J.H. Connell and R.M. Bolle,
Conference, Sept. 4-8, IEEE Xplore Press, Antalya, 2007. Generating cancelable fingerprint templates.
Turkey, pp: 1-4. IEEE Trans. Pattern Analysis Machine Intelligence,
29: 561-572. DOI: 10.1109/TPAMI.2007.1004
Jain, A.K., K. Nandakumar and A. Nagar, 2008. Biometric Šošević, U., I. Milenković, M. Milovanović and M.
template security. EURASIP J. Adv. Signal Process., Minović, 2013. Support platform for learning about
2008: 1-17. DOI: 10.1155/2008/579416 multimodal biometrics. J. Universal Comput. Sci.,
Jain, A.K., S. Prabhakar and A. Ross, 1998. Biometric- 19: 1684-1700.
based web access. Trans. Institute Brit. Geographers. Teodoro, N. and C. Serrao, 2011. Web application
Lee, H. and R. Gaensslen, 2001. Advances in Fingerprint security: Improving critical web-based applications
Recognition. 2nd Edn., S.l.: CRC Press, Taylor and quality through in-depth security analysis.
Francis Group. Proceedings of the International Conference on
Information Society (i-Society), Jun. 27-29, IEEE
Merati, A. 2011. Multi-modal biometric authentication
Xplore Press, London, UK.
with cohort-based normalization. PhD Theses, Zhang, D., F. Song and Y. Xu, 2009. Advanced Pattern
Centre for Vision, Speech and Signal Processing, Recognition Technologies with Applications to
Faculty of Engineering and Physical Sciences, Biometrics. 1st Edn., Book News Inc., ISBN-10:
University of Surrey, UK. 1605662003, pp: 366.

693

View publication stats