Docstoc

Combining Level- 1 ,2 & 3 Classifiers For Fingerprint Recognition System

Document Sample
Combining Level- 1 ,2 & 3 Classifiers For Fingerprint Recognition System Powered By Docstoc
					                                          (IJCSIS) International Journal of Computer Science and Information Security,
                                          Vol. 8, No. 7, October 2010




 COMBINING LEVEL- 1 ,2 & 3 CLASSIFIERS FOR
   FINGERPRINT RECOGNITION SYSTEM

      Dr.R.Seshadri ,B.Tech,M.E,Ph.D                     Yaswanth Kumar.Avulapati,M.C.A,M.Tech,(Ph.D)
      Director, S.V.U.Computer Center                     Research Scholar, Dept of Computer Science
          S.V.University, Tirupati                                  S.V.University, Tirupati
     E-mail : ravalaseshadri@gmail.com                      E-mail:yaswanthkumar_1817@yahoo.co.in



Abstract                                                 Keywords-Biometrics,         Classifier,Level-
                                                         1,Level-2 features,Level-3 features
       Biometrics is the science of
establishing the identity of an person based             Introduction
on their physical, chemical and behavioral
characteristics of the person. Fingerprints are                 A fingerprint is a pattern of ridges and
the most widely used biometric feature for               valleys located on the tip of each finger.
person identification and verification in the            Fingerprints were used for personal
field of biometric identification .A finger              identification for many centuries and the
print is the representation of the epidermis of          matching accuracy was very high. Human
a finger. It consists of a pattern of interleaved        fingerprint recognition has a tremendous
ridges and valleys.                                      potential in a wide variety of forensic,
       Fingerprints are graphical flow-like              commercial       and      law      enforcement
ridges present on human fingers. They are                applications.
fully formed at about seven months of fetus
development and finger ridge configurations                     Fingerprints are broadly classified into
do not change throughout the life of an                  three levels they are Level-1 which includes
individual except due to accidents such as               arch,tentarch, loop, double Loop, pocked
bruises and cuts on the fingertips.                      Loop, whorl ,mixed, left-loop, right-loop the
                                                         Level-2 includes the minutiae and Level 3
       This property makes fingerprints a                includes pores etc.
very attractive biometric identifier. Now a                     There are so many approaches are
day’s fingerprints are widely used among                 there for recognizing the fingerprints among
different biometrics    technologies. In this            these correlation based, minutiae based, ridge
paper we proposed         an approach to                 feature based are most popular ones.
classifying the fingerprints into different                     Several biometrics systems have been
groups. These fingerprints classifiers are               successfully developed and installed. How
combined together for recognizing the people             ever some methods do not perform well in
in an effective way.                                     many real-world situations due to its noise.




                                                   128                               http://sites.google.com/site/ijcsis/
                                                                                     ISSN 1947-5500
                                       (IJCSIS) International Journal of Computer Science and Information Security,
                                       Vol. 8, No. 7, October 2010



                                                               ,


Fingerprint Classifier

Here we proposed a fingerprint classifier
framework. A combination scheme involving
different fingerprint classifiers which
integrates vital information is likely to
improve the overall system performance.

The fingerprint classifier combination can be
implemented at two levels feature level and
decision level. We use the decision level
combination that is more appropriate when                     Fig 1.       Fingerprint Level 1 Features
the component classifiers use different types
of features. Kittler provides a theoretical                  Level 2 features describe various ridge
framework to combine various classifiers at           path deviations where single or multiple
the decision level. Many practical                    ridges form abrupt stops, splits, spurs
applications of combining multiple classifiers        bifurcation Composite minutiae (i.e., forks,
have been developed. Brunelli and Falavigna           spurs, bridges, crossovers and bifur-cations)
presented a person identification system by           can all be considered as combinations of
combining outputs from classifiers based on           these basic forms enclosures, etc. These
Audio and visual.                                     features, known as the Galton points or
                                                      minutiae, have two basic forms: ridge ending
Here the combination approach is designed at          and ridge as shown in fig 2.
the decision level utilizing all the available
information, i.e. a subset of (Fingerprint)
labels along with a confidence value, called
the matching score provided by each of the
nine finger print recognition method.

Classification of Fingerprint
(Level-1,Level -2 & Level-3) Features                              Fig 2. Fingerprint Level 2 Features

                                                            Level 3 features refer to all
       Level 1 features describe the ridge
                                                      dimensional attributes of a ridge, such as
flow pattern of a fingerprint. According to
                                                      ridge path deviation, width, shape, pores,
the Henry classification system there are
                                                      edge contour, incipient ridges, breaks,
eight major pattern classes, comprised of
                                                      creases, scars and other permanent details as
whorl, left loop, right loop, twin loop, arch,
                                                      shown in fig 3.
tented arch. as shown in the figure 1.



                                                129                               http://sites.google.com/site/ijcsis/
                                                                                  ISSN 1947-5500
                                          (IJCSIS) International Journal of Computer Science and Information Security,
                                          Vol. 8, No. 7, October 2010




                                                                                      Level -1                          Matching
                                                                                      Level -2                           Score
                                                                                       Level-3
                                                                                     Features of
                                                                                    finger prints

                                                         Fingerprint

                                                                                                                      Final Out Put


                                                                                                                Final Out Put




                                                        Training Fingerprint
  Fig 3. Fingerprint Level 3 Features
                                                        Fig 4.Fingerprint Classifier Combination
Classifier Combination System
                                                        System
       We proposed a classifier combination
                                                        Combination Strategy
shown in the Fig .Here currently we use only
nine classifiers     for level-1 features of
                                                              Kittler analyzed several classifier
fingerprints namely arch,tentarch, loop,
                                                        combination rules and concluded that the
double Loop, pocked Loop, whorl ,mixed,
                                                        sum rule as shown in the given below
left-loop, right-loop
                                                        outperforms other combination schemes
                                                        based on empirical observations.
      For Finger print level-2 features
namely right-loop various ridge path
                                                               Unlike       explicitly     setting      up
deviations where single or multiple ridges
                                                        combination rules, it is possible to design a
form abrupt stops, splits, spurs bifurcation
                                                        new classifier using the outputs of individual
Composite minutiae (i.e., forks, spurs,
                                                        classifiers as features to this new classifier.
bridges, crossovers and bifurcations
                                                               Here we assume the RBF network as a
       For Level-3 features namely deviation,
                                                        new classifier. Given m templates in the
width, shape, pores, edge contour, incipient
                                                        training set, m matching scores will be output
ridges, breaks, creases, scars
                                                        for each test image from each classifier.
                                                        We consider the following two integration
        Following two strategies are provided
                                                        strategies
for integrating outputs of individual
classifiers, (i) the sum rule, and (ii) a RBF
                                                130                                  http://sites.google.com/site/ijcsis/
network as a classifier, using matching scores                                       ISSN 1947-5500
as the input feature vectors as shown in fig 4.
                                          (IJCSIS) International Journal of Computer Science and Information Security,
                                          Vol. 8, No. 7, October 2010



                                                            Level-1      Level-2       Level-3
     1. Strategy I: Sum Rule. The combined                  Features     Features      Features
matching score is calculated as                                70             75            90
                                                                Fig.5a recognition accuracies of
    Macomb = MSPCA +MSICA +MSLDA:                        different    finger     print  recognition
                                                         approaches are listed
       For a given sample,Output the class
with the largest value of Macomb.
                                                                       Cumulative match score vs. rank
                                                                 100      curve for the sum rule.
        2. Strategy II: RBF network. For each                     90
test image, the m matching scores obtained
                                                                  80
from each classifier are used as a feature
                                                                  70
vector. Concatenating these feature vectors
derived       from      Level-1,Level-2,Level-3                   60




                                                          Rank
classifiers results in a feature vector of size                   50
3m.                                                               40
        An RBF network is designed to use                         30
this new feature vector as the input to                           20                                                     Series1
generate classification results. We adopt a                       10
Level-1,Level-2,Level-3         layers      RBF
                                                                   0
network. The input layer has 3 levels m                                  Level 1      Level 2        Level 3
nodes and the output has c nodes, where c is                            Features     Features       Features
the total number of classes (number of                                      Cumulative Match Score
distinct features of fingerprints). In the output
layer, the class corresponding to the node
with the maximum output is assigned to the               Figure 5 b show that the combined
input image. The number of nodes in the                  classifiers, based on both the sum-rule and
hidden layer is constructed empirically,                 RBF network, outperform each individual
depending on the sizes of the input and                  classifier.
output layers. Sum score is output as the final
result.                                                  CONCLUSION

       The recognition accuracies of different                 Finally we conclude that in our
finger print recognition approaches are listed           proposed approach the combination scheme
in table 5a. The cumulative match score vs.              which combines the output matching scores
rank curve is used to show the performance               of three levels of well-known Fingerprint
of each classifier, see Fig 5b. Since our RBF            recognition system. Basically we proposed
network outputs the final label, no rank                 the model to improve the performance of a
information is available. As a result, we                fingerprint identification system at the same
cannot compute the cumulative match score                time the system provides high security from
vs. rank curve for RBF combination                       unauthorized access.

                                                                Two mixing strategies, sum rule and
                                                         RBF-based integration are implemented to
                                                   131   combine the output information of three level
                                                                            http://sites.google.com/site/ijcsis/
                                                                            ISSN 1947-5500
                                                         features of fingerprint 0individual classifiers.
                                       (IJCSIS) International Journal of Computer Science and Information Security,
                                       Vol. 8, No. 7, October 2010



The proposed system framework is scalable            ment of Computer Science and Engineering,
other fingerprint recognition modules can be         Michigan State University, 2008.
easily added into this framework. Results are
encouraging, illustrating that both the              [9].K. Kryszczuk, A. Drygajlo, and P.
combination strategies lead to more accurate         Morier. Extraction of Level 2 and Level 3
fingerprint recognition than that made by any        Features for Fragmentary Fingerprints. In
one of the individual classifiers.                   Proc. COST Action 275 Workshop,
                                                     pages 83{88, Vigo, Spain, 2004.
References
                                                     [10]A. K. Jain, S. Prabhakar, and S. Chen,
[1].A. K. Jain,Patrick Flynn,Arun A.Ross .           “Combining multiple matchers for a high
“Handbook of Biometrics”.                            security fingerprint verification system,”
                                                     Pattern Recognition Letters, vol. 20, no. 11-
[2].D. Maltoni, D. Maio, A. K. Jain, and S.          13, pp. 1371–1379, 1999.
Prabhakar,    Handbook      of  Fingerprint
Recognition. Springer, 2003.                         Authors Profile
                                                                             Dr.R.Seshadri was born in
[3.] N. Yager and A. Amin. Fingerprint                                       Andhra Pradesh, India, in
                                                                             1959. He received his
classi_cation: A review. Pattern Analysis
                                                                             B.Tech      degree       from
Application, 7:77{93, 2004.                                                  Nagarjuna University in
                                                                             1981. He received his M.E
[4]. O. Yang, W. Tobler, J. Snyder, and Q. H.                                degree in Control System
Yang. Map Projection Transforma-tion.                                        Engineering from PSG
Taylor & Francis, 2000.                                                      College of Technology,
                                                                             Coimbatore in 1984. He was
                                                     awarded with PhD from Sri Venkateswara
[5]. Z. Zhang. Flexible Camera Calibration           University, Tirupati in 1998. He is currently
by Viewing A Plane from Unknown                      Director,    Computer     Center,    S.V.University,
Orientations. IEEE Transactions on Pattern           Tirupati, India. He has Published number of papers
Analysis            and           Machine            in national and international conferences, seminars
Intelligence,11:1330{1334, 2000.                     and journals. At present 12 members are doing
                                                     research work under his guidance in different areas
[6]. J. Zhou, C. Wu, and D. Zhang.                                                      Mr.YaswanthKumar
Improving Fingerprint Recognition Based on                                    .Avulapati received his
Crease Detection. In Proc. International                                      MCA degree with First
Conference on Biometric Authentication                                        class from Sri Venkateswara
(ICBA), pages 287{293, Hong Kong, China,                                      University, Tirupati. He
                                                                              received          his          M.Tech
July 2004.
                                                                              Computer           Science        and
                                                                              Engineering degree with
[7]. Y. Zhu, S. Dass, and A. K. Jain.                                         Distinction from Acharya
Statistical Models for Assessing the                                          Nagarjuna                 University,
Individual- ity of Fingerprints. IEEE                                         Guntur.He is a research
Transactions on Information Forensics and                                     scholar in S.V.University
                                                     Tirupati, Andhra Pradesh.He has presented number
Security, 2:391{401, 2007.
                                                     of papers in national and international conferences,
                                                     seminars.He attend Number of work shops in
[8]. Y. F. Zhu. Statistical Models for132            different fields.
                                                                          http://sites.google.com/site/ijcsis/
                                                                          ISSN 1947-5500
Fingerprint Individuality. PhD thesis, Depart-

				
DOCUMENT INFO
Description: Vol. 8 No. 6 September 2010 International Journal of Computer Science and Information Security