Docstoc

FINGERPRINT IDENTIFICATION TECHNIQUE

Document Sample
FINGERPRINT IDENTIFICATION TECHNIQUE Powered By Docstoc
					  International Journal of JOURNAL OF COMPUTER (IJCET), ISSN 0976-
 INTERNATIONALComputer Engineering and Technology ENGINEERING
  6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
                          & TECHNOLOGY (IJCET)

ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online)                                                    IJCET
Volume 4, Issue 3, May-June (2013), pp. 308-323
© IAEME: www.iaeme.com/ijcet.asp
Journal Impact Factor (2013): 6.1302 (Calculated by GISI)
                                                                         ©IAEME
www.jifactor.com




           FINGERPRINT IDENTIFICATION TECHNIQUE
     BASED ON WAVELET-BANDS SELECTION FEATURES (WBSF)

                               Dr. Mustafa Dhiaa Al-Hassani,
                             Mustansiriyah University, Baghdad-Iraq

                                Dr. Abdulkareem A. Kadhim,
                              Al-Nahrain University, Baghdad-Iraq

                                    Dr. Venus W. Samawi,
                                  Al al-Bayt University, Jordan



  ABSTRACT

          The paper is concerned with the use of fingerprint (FP)features for protection against
  unauthorized access. Wavelet features for both closed and open-set FP recognition are studied
  here to verify persons' identity. Fingerprints of 49 persons (32-authorized and 17-unauthorized)
  were taken as testing data. Each authorized person is asked to give 10-instances of his right
  forefinger print. In the closed-set FP recognition, the obtained recognition rates are below 90%
  due to the imperfections in the FP images that negatively affect the recognition rate.
  Preprocessing operations such as: noise-removal, segmentation, normalization and binarization
  are considered to improve the resulting recognition rates. A method that relies on a new
  selection process for wavelet decomposition bands is proposed, which enhance the recognition
  rates further to get about 100% in some favorable conditions. The results have shown that the
  wavelet descriptors using the proposed Wavelet-Bands Selection Features (WBSF) are efficient
  representation that can provide reliable recognition for large input variability. The open-set FP
  verification mode is also presented for 290 trials from 29 persons, where the obtained
  verification rates are greater than 97% for both Euclidean and city-block distance measures.

  Keywords: Fingerprint Recognition, Fingerprint Verification, Biometric, Feature Extraction,
  Wavelet Transform, Wavelet-Bands Selection Features.



                                               308
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME

I. INTRODUCTION

        Due to the escalating level of security breaches and transactions frauds, the need for
highly secure identification and personal verification technologies becomes essential. The key
task of an automated security system is to verify that the users are in fact those who claim to
be. The use of biometric information has been used widely for both person identification and
security applications. Biometric-based solutions are able to provide confidential financial
transactions and personal data privacy [1]. A biometric can be described as a measurable
physical and/or behavioral trait that can be captured and used to verify the identity of a person
[2].
        FP recognition is a rapidly evolving technology that has been widely used in forensics
such as criminal recognition and prison security, and has widely adopted in a broad range of
civilian applications such as national ID card, airport check-in, border control, driver’s-license
authenticity, computer network logon, physical access control, electronic banking, personal
authentication,… etc [3, 4, 5].
        The real significance of FP is based mainly on the following principles: 1) People can't
"forget" their fingerprints, 2) It is easy to authenticate, 3) Impossible to deny, 4) It is a physical
characteristic instead of something to be remembered or carried around; it is less susceptible to
misuse than other authentication measures like passwords or credit cards, 5) Unchangeable, 6)
Unforgeable, ... etc [6, 7].
        Several researches in the field of FP recognition/verification were developed and
receive a great deal of attention among many researchers using of wavelet transform and other
feature extraction methods: Priti and Priyadarshan [4] introduced a FP verification using Haar
wavelet transform method. The system was tested on a Biolab Database of 2160 FP images.
The obtained verification accuracy is 82.08% even by rotating each FP image from 00 to 3600.
Eriksson [7] illustrated that silicon FP scanners produce good quality images, this work
presents two main approaches to minutia detection in FP images, binary detection and direct
grayscale detection. The results are tested on 6283 fingerprints collected by the Verdicom
FPS110 silicon FP scanning device and they reported about 92% classification accuracy. Saeed,
Tariq, and Jawaid [8] improved a fingerprint image enhancement technique using Gabor
wavelets. The system was tested on Fingerprint Verification Competition (FVC) 2004 database.
Experimental results show that the proposed algorithm proved to be effective in enhancing the
fingerprint image quality, where the achieved accuracy is 97.14%. In [9] the authors presented
minutiae based approach to FP identification and verification. The technique is based on the
extraction of minutiae from the thinned, binarized and segmented version of a FP image. The
system was tested on the FVC2000 database using low cost capacitative FP scanners, which
contains 800 fingerprints from 110 different fingers. The system was implemented using
Matlab 6.5 and the time taken for processing a single FP is 12 seconds that implies accuracy
92%.

II. AIM OF THE WORK

         This work aims to design and build a secure, fast, reliable, and accurate identification
system for access control that is capable of distinguishing the authorized persons from others
(i.e., impostors), and then gives only the authorized persons a privilege or an access right to the
facility that need to be protected from the intrusion of unauthorized persons.


                                                309
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME

    In this paper, a novel Wavelet feature-set (WBSF) is proposed for representing FP pattern.
FP recognition and verification is also to be investigated for open and closed-set models.

III. THE PROPOSED FP RECOGNITION SYSTEM MODEL

        In general, the function of FP systems can be separated into several distinct phases,
which include sensing or reading FP, preprocessing operations, FP registration, feature
extraction followed by a classification search and decision rule [1, 10]. The block diagram for
the proposed FP recognition system model, shown in Fig. (1), illustrates that the input FP
image is passed through four preprocessing operations (noise-removal, segmentation,
normalization and binarization) prior to feature extraction phase [1].




                 Fig. (1): Block-Diagram of the proposed FP Recognition
                                      System Model

        Features are extracted from wavelet domain, using the classical pyramidal Wavelet
transform decomposition followed by the features extracted from the proposed Wavelet-Bands
Selection Features (WBSF), as shown in the design of the proposed system to recognize a
query FP image by comparing it with a training database of F Preferences during a pattern
matching phase. Finally, the distance measures (Euclidean and City-block) are used to calculate
the difference between the feature vector of the query FP with the feature vector of the
potential FP in the database. The next subsections will cover the details of each stage [1].

A. Input FP Image

     Figure (2) illustrates some examples of input FP images used for training or testing modes
to our system model from the right forefinger of different persons (P1, P2, …, P6) [1].



                                            310
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME




                         P1     P2        P3




                         P4             P5       P6

                  Fig.(2):Example of 6–FP images from different Persons

B. Preprocessing of the FP Images

         FP images are rarely of perfect quality. They may be degraded and corrupted with
elements of noise due to many factors including variations in skin and impression conditions. It
must also overcome fingers pressed too hard or too gently to get an acceptable image. Getting
an acceptable image is probably the most important factor in determining fingerprints
genuineness. Bad quality prints can result in unsuccessful recognition attempts or even worse,
erroneous logins. Thus, image enhancement techniques are employed prior to feature extraction
to reduce the noise and enhance the definition of ridges against valleys. A number of
processing techniques adopted in this system model are applied in the following sequence [1]:
Noise-Removal (using Mean or Gaussian filter), Segmentation (foreground/background
separation), Normalization (to reduce the effect of non-uniform intensities and improving
image quality by stretching its histogram), and Binarization (using local mean). Figure (3)
illustrates the effects on a sample FP image [1].




             Fig. (3): The sequence of preprocessing steps for FP image sample




                                               311
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME

C. Database Construction

      FP identification system (both FP recognition and FP verification) depends on FP samples
as input data. In this work, database samples were collected in two modes of operation:
   • Closed-set FP recognition mode
   • Open-set FP verification mode
      In order to evaluate the recognition performance of the proposed system model, each user
of the system (to be considered as authorized one) has been asked to provide his forefinger
print for a maximum of 10 prints from the same forefinger (i.e., 10 instances), as shown in Fig.
(4). The number of repetition R ( 1≤ R ≤ 10 )can be considered as training set during an
enrollment phase to train the fingerprints model of authorized persons, and the other ( 10 – R )
repetitions are considered for testing during a matching phase to classify them with those
training references in the database [1].




                                    P2,1 P2,2            P2,3




                                P2,4     P2,5            P2,6


                 Fig. (4): Demonstrates 6-FP instances from the same Person P2

The data were collected from 32 different persons, 18 males and 14 females, in a closed set FP
recognition mode (i.e. 320 samples). As a result, the total database size of FP samples for this
mode is [1]:

   Total DB S ize = 10 × No. of Pe rsons                                               …… (1)

   No . of Training References = R × No . of Persons                                   …… (2)

   No. of Test Samples = (10 − R) × No . of Persons                                    …… (3)

In the open-set FP verification mode, up to 290 trials from different persons (i.e., authorized
and unauthorized) are taken. This is performed in order to study the system behavior and to
select the optimal threshold for user verification.

D. Feature Extraction

       The process of extracting some numerical measurements from raw input patterns by
constructing a new "smaller" set of features from the original feature set of patterns (i.e.


                                                       312
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME

reducing dimensionality) is referred to as feature extraction[11].In this work, features are
extracted from the spectral properties of the wavelet transform.
        Wavelet transform breaks an image down into four sub-sampled images, and then
analyze each component with a resolution matched to its scale. The forward and inverse
continuous wavelet transform ofx(t) with respect to the basis function or wavelet ψ j , k (t) at scale j
(j>0) and time delay k is written as follows [4, 12 –14]:

     Forward CWT : W ( j, k ) = ∫ x ( t ) ψ j, k ( t ) dt                                       ..…(4)

     Inverse CWT : x(t ) =   ∫∫ W( j, k) ψ j, k (t) dk dj                                       ..…(5)
                             k j
where

                   ψ j , k (t) = 1 ψ ( t − k )                                                  ..…(6)
                                    j        j



and ψ (t) is the mother wavelet.
After converting the input FP image from its lowest-level of pixel data into higher-level
representation of wavelet coefficients, [1],a set of wavelet features that represent the input FP
image can be extracted by recursively decomposing sub images in the low frequency channels
using Algorithm-1 as shown below:

     Algorithm-1:The Classical Pyramidal Wavelet Transform Decomposition [1, 15]

Step1: Decompose a given textured image with 2-D wavelet transform into 4 sub images, as
indicated in Fig. (5) (the image is divided into four sub bands after wavelet transform:
horizontal, vertical, diagonal subimages and low resolution subimage).


                                         LL3     LH3
                                                        LH2
                                         HL3 HH3
                                                                  LH1

                                           HL2          HH2




                                                  HL1             HH1




                                Fig. (5): Three-level Wavelet Decomposition



                                                            313
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME

Step2: 1) Calculate the Mean Absolute Value (M.A.V.) feature for each decomposed image, as
        follows [15]:
                                        M     N
                             1
         M.A.V. =
                     M×N
                                       ∑∑         x ( i , j)                               …… (7)
                                       i =1 j=1


where
   x(i, j) is the decomposed image, with 1 ≤ i ≤ M and 1 ≤ j ≤ N . M is the subimage height
  and N is the subimage width.
2) Calculate the Standard Deviation (S.D.) feature for x(i, j) , as shown below [15, 16]:
                                 M     N
                     1
          S.D. =
                    M ×N
                                 ∑ ∑ ( x (i , j ) − M.A.V. )
                                 i = 1 j =1
                                                               2
                                                                                           …… (8)


The size of the smallest subimages should be used as a stopping condition for the iterative
decomposition process. It is also worthwhile to point out that the above pyramidal wavelet
transform decomposition takes no more space to store the wavelet coefficients than it does to
store the original image.

E. Pattern Matching

       The resulting test template, which is an N-dimensional feature vector, is compared
against the stored reference templates to find the closest match. The process is to find which
unknown class matches a predefined class or classes. For the FP recognition task, the unknown
FP is compared to all references in the database. This comparison can be done through
Euclidean (E.D.) or city-block (C.D.) distance measures [17], shown below:

                         N
         E .D . =     ∑ (a i − b i ) 2                                                     …… (9)
                      i =1


                     N                                                                     ….. (10)
         C .D . =    ∑           a i − bi
                     i =1

where A and B are two vectors, such that A = [a1 a2 … aN]and B = [b1 b2 … bN].
        The primary methods for the discrimination process are either to measure the difference
between the two feature vectors or to measure the similarity. In our approach the minimum
distance classifier, by measuring the difference between the two patterns, is used for FP
recognition. This classifier assigns the unknown pattern to the nearest predefined pattern. The
bigger distance between the two vectors, is the greater difference. On the other hand, the
identity of the unknown FP was verified by considering the best matched reference in the
database where their distance is lower than a certain threshold [17, 18].

IV. EXPERIMENTAL RESULTS

        The recognition rate (R.R.) is defined as the ratio of correct identified fingerprints to the
total number of test samples which corresponds to a nearest neighbor decision rule.


                                                                   314
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME

              No. of Correctly Identified FP
     R.R. =                                  × 100 %                                          ..… (11)
              Total No. of Samples Tested

        Many experiments and test conditions were accomplished to measure the performance
of the proposed system with different criterions concerning: wavelet decomposition level
selection, FP noise-removal, segmentation, normalization, binarization, the effect of the
proposed WBSF on the overall recognition/verification rates when compared with the classical
pyramidal wavelet transform decomposition method.

A. The Selection of Wavelet Level

       In order to select the best level of wavelet decomposition for the system, this test is
performed. Different Daubechies wavelet functions are considered as shown in Table-1.

 Table-1: Recognition rates for different levels of wavelet decomposition using (M.A.V.)
                                         feature


              Wavelet
                               Level-1          Level-2         Level-3   Level-4   Level-5
              Function
                 D2             41.875           55.000         75.000    76.875    76.875

                 D4             49.375           60.625         79.375    84.375    77.500

                 D6             46.250           65.000         82.500    84.375    83.750

                 D8             41.875           64.375         84.375    86.250    83.125

                D10             43.125           65.000         84.375    87.500    86.875

                D12             45.625           65.000         84.375    89.375    85.000

                D14             44.375           63.750         84.375    87.500    85.000

                D16             43.750           65.000         82.500    83.750    85.625

                D18             44.375           65.000         85.000    86.875    86.250

                D20             41.875           66.875         84.375    87.500    82.500



       The wavelet levels considered for each function is varied from 1 to 5. The number of
wavelet features for the first level is 4, and each progressing in wavelet level by iteratively
decomposing the low resolution sub image will correspond increasing in features length by 3.
For each training or testing, five repetitions for each FP are considered which resulted in 160
samples.
       It is clear from Table-1 and its corresponding chart Fig. (6), that Level-4 is the most
appropriate level for feature vector construction where all the recognition ratesare the highest
among almost all Daubechies functions.

                                                          315
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME


                                                                             Level 1            Level2            Level3
                                                                             Level 4            Level5
                                                             100




                                            Recognition Rate %
                                                                 90

                                                                 80

                                                                 70
                                                                 60

                                                                 50

                                                                 40
                                                                       D2    D4 D6 D8           0  2  4  6  8
                                                                                              D1 D1 D1 D1 D1 D20
                                                                             Daubechies Wavelet functions


 Fig. (6): Recognition rates for different levels of wavelet decomposition using (M.A.V.)
                                          feature

B. Segmentation of Low-pass Filtered FP images

     After determining the appropriate wavelet decomposition level, a number of preprocessing
steps were performed to enhance image quality. The first step is to remove the noise from the
input FP images using the Gaussian-filter, and then separating the foreground regions from the
background regions in a FP image. Figure(7) shows the resulting recognition rates when
Euclidean distance measure and M.A.V. were used.

                                                                            Gaussian Filter              with Segmentatio n

                                            100

                                                    95
                       Recognition Rate %




                                                    90

                                                    85

                                                    80

                                                    75

                                                    70

                                                    65
                                                                      D2 D4 D6 D8 D10 D12 D14 D16 D18 D20
                                                                               Daubechies Wavelet functions

    Fig.(7): Effects of Segmentation on recognition rates for the Gaussian-filtered FP


        It is clear from Fig. (7), that the segmentation process enhances all the recognition rates
for the Gaussian-filtered FP images; where about (96%) recognition rate is achieved using D8.
On the other hand, when segmentation is not used, all recognition rates are below 90%.




                                                                                              316
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME

C. Binarization of Normalized FP images

    Further improvements in recognition rates can be achieved when converting the
normalized FP images (using histogram stretching) from gray-scale to binary as shown in
Table-2. This increases the contrast among the ridges and valleys of FP.

    Table-2: Recognition rates after Binarization of the Normalized FP images using
                                 (M.A.V. and S.D.) features

          Wavelet                                           E. D.                                        C. D.
          Function                              M.A.V.                S.D.               M.A.V.                    S.D.
             D2                                 90.000               96.250              87.500                  94.375
             D4                                 95.625               97.500              95.625                  97.500
             D6                                 96.250               99.375              96.875                  98.750
             D8                                 98.750               100.00              98.750                  99.375
            D10                                 95.625               100.00              96.875                  98.750
            D12                                 99.375               99.375              97.500                  99.375
            D14                                 96.250               99.375              96.875                  100.00
            D16                                 98.125               100.00              98.125                  100.00
            D18                                 98.750               100.00              98.750                  99.375
            D20                                 98.125               99.375              100.00                  99.375

        The results of Table-2 obviously indicate the highly enhancements in all recognition
rates after applying the binarization step to the normalized FP images for both distance
measures when compared to previous test. Furthermore, one can deduce that the (S.D.) wavelet
feature present better results than (M.A.V.) feature using both distance measures. Figure (8)
display part of this comparison by taking only the (S.D.) wavelet feature using Euclidean
distance measure.


                                                       No rmalizatio n             With B inarizatio n

                                          100
                                           98
                                           96
                                           94
                     Recognition Rate %




                                           92
                                           90
                                           88
                                           86
                                           84
                                           82
                                           80
                                           78
                                           76
                                                  D2   D4    D6   D8     D1 D1
                                                                           0  2     4  6  8
                                                                                  D1 D1 D1 D20
                                                            Daubechies Wavelet functions


Fig. (8): Recognition rates for wavelet feature (S.D.) after Binarization of Normalized FP
                                          images

                                                                         317
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME

D. The Proposed Wavelet–Bands Selection Features (WBSF)

        This section reviews a novel selection method for the set of wavelet features that are
well suited for recognition of FP images with the aim to improve the recognition rates. In this
method, the wavelet features extracted by means of four wavelet decomposition levels (i.e. 13
features) are combined with another (18 features) extracted from five decomposition levels
wavelet bands as shown in Fig. (9). These provide information about FP image in both
horizontal and vertical directions [1].The added features are the shaded cells shown in Fig.
(10).The final calculated 31 features are arranged in a single vector that will represent the FP
feature pattern.




   Fig.(9): Demonstrates 5–decomposition levels of the 2-D wavelet transform for a FP
                                       sample




    Fig.(10): The proposed wavelet channels decomposition (5-levels) by indicating the
                          number of each newly selected band


                                             318
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME

        The experimental results for the proposed WBSF system are shown in Table-3with its
corresponding Figures (11) and (12), which gives better recognition rates when compared to
those in Table-2.This is due to the fact that the proposed WBSF method for feature extraction
provides extra information to assist further in recognition.


    Table-3: Recognition rates for WBSF system with different wavelet functions and
                                    distance measures

         Wavelet                                          E. D.                                            C. D.
         Function                               M.A.V.                 S.D.                 M.A.V.                 S.D.
             D2                                 95.000              98.125                   93.125                97.500
             D4                                 97.500              99.375                   96.250                98.125
             D6                                 98.125              100.00                   98.125                99.375
             D8                                 98.750              100.00                   97.500                100.00
             D10                                100.00              100.00                   100.00                99.375
             D12                                100.00              100.00                   100.00                100.00
             D14                                100.00              100.00                   99.375                100.00
             D16                                98.125              100.00                   98.750                100.00
             D18                                100.00              100.00                   100.00                100.00
             D20                                100.00              100.00                   100.00                99.375




                                                        With Binarization                      WBSF
                                          100
                                           99
                                           98
                                           97
                     Recognition Rate %




                                           96
                                           95
                                           94
                                           93
                                           92
                                           91
                                           90
                                           89
                                           88
                                                  D2    D4   D6   D8   D 10   D 12 D 14 D 16 D 18   D 20
                                                       Daubechies Wavelet functions



Fig. (11): Recognition rates for wavelet feature (M.A.V.) before and after the addition of
                                   the proposed WBSF




                                                                       319
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME


                                                       With Binarization                    WBSF
                                        100
                                            99




                       Recognition Rate %
                                            98
                                            97
                                            96
                                            95
                                            94
                                            93
                                            92
                                            91
                                            90
                                            89
                                            88
                                                 D2    D4   D6   D8   D 10 D 12 D 14 D 16 D 18 D 20

                                                      Daubechies Wavelet functions

Fig. (12): Recognition rates for wavelet feature (S.D.) before and after the addition of the
                                     proposed WBSF

      In WBSF, the decomposition bands involved are not only the LL bands but also the LH
and HL bands that correspond to horizontal and vertical FP image details, respectively.
E. Fingerprint Verification
     The final step requires the verification of user’s identity. This is relies on the best results
obtained from the previous experiments. It is undoubtedly illustrated that (S.D.) based feature,
extracted from different wavelet bands, exhibits better results when compared to(M.A.V.).
Therefore, (S.D.) is selected to be the feature for wavelet extraction method of the FP
verification mode using WBSF features set.
   Since different wavelet functions can provide recognition rates close to 100% as illustrated
in Table-3, therefore we select D20 as the wavelet function for the verification tests. A total of
290 query FP samples from 29 persons (authorized and unauthorized) are considered for open-
set FP verification mode. Different threshold values were considered, as shown in Table-4 and
5. The successful decision corresponds to the rate of accepting registered persons and rejecting
non-registered ones for all trials.

              Table-4: FP verification rates for D20 using Euclidean distance

           Threshold (θ)                         Successful Decision                  FAR             FRR
               2.60                                     73.4482                        0.0            26.5517
               2.85                                     78.2758                        0.0            21.7241
               3.10                                     83.1034                        0.0            16.8965
               3.35                                     91.7241                        0.0             8.2758
               3.60                                     94.1379                        0.0             5.8620
               3.85                                     97.2413                      0.6896            2.0689
               4.10                                     96.5517                      2.4137            1.0344
               4.35                                     95.1724                     3.7931            1.0344
               4.60                                     92.4138                      6.8965            0.6896
               4.85                                     89.3103                     10.0000            0.6896
               5.10                                     86.5517                     12.7586            0.6896
               5.35                                     82.4138                     16.8965            0.6896
               5.60                                     78.6206                     21.0344            0.3448

                                                                       320
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME

              Table-5: FP verification rates for D20 using city-block distance

                                            Successful Decision                                       FAR                       FRR
        Threshold (θ)
            12.00                                        80.6896                                        0.0                     19.3103
            12.50                                        85.1724                                        0.0                     14.8275
            13.00                                        88.2758                                        0.0                     11.7241
            13.50                                        91.7241                                        0.0                     8.2758
            14.00                                        94.1379                                        0.0                     5.8620
            14.50                                        96.5517                                        0.0                     3.4482
            15.00                                        97.9310                                        0.0                     2.0689
            15.50                                        97.9310                                        0.0                     2.0689
            16.00                                        97.5862                                      0.3448                    2.0689
            16.50                                        97.5862                                      1.3793                    1.0344
            17.00                                        95.8620                                      3.1034                    1.0344
            17.50                                        95.1724                                      4.1379                    0.6896
            18.00                                        93.4482                                      5.8620                    0.6896
            18.50                                        92.0689                                      7.2413                    0.6896
            19.00                                        89.6551                                      9.6551                    0.6896


        The optimum threshold of Crossover Error Rate (CER) is the point where the False
Rejection Rate (FRR) and the False Acceptance Rate (FAR) curves meet in verifying user's
identity. The variation of FAR and FRR with different threshold values are also shown in Fig.
(13) and (14),where the obtained CER are approximately 4.03 and 16.45 for Euclidean and
city-block distances respectively.


                                                               FAR                                     FRR
                                       30
                                       28
                                       26
                                       24
                                       22
                        Error Rate %




                                       20
                                       18
                                       16
                                       14
                                       12
                                       10
                                        8
                                        6
                                        4
                                        2
                                        0

                                            2.6   2.85   3.1   3.35   3.6   3.85   4.1   4.35   4.6   4.85   5.1   5.35   5.6

                                                    Threshold values for Euclidean distance

     Fig. (13): FAR and FRR Performance Curve for different threshold levels using
                                 Euclidean distance




                                                                              321
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME


                                                               FAR                                          FRR
                                      22
                                      20

                                      18




                       Error Rate %
                                      16

                                      14

                                      12

                                      10

                                      8
                                      6
                                      4

                                      2

                                      0
                                           12   12.5   13   13.5   14   14.5   15   15.5   16   16.5   17   17.5   18   18.5   19

                                                  Threshold values for City-block distance



  Fig. (14): FAR and FRR Performance Curve for different threshold levels using city-
                                  block distance

V. CONCLUSIONS

        A fingerprint recognition system for 490 FP samples is presented that relies on wavelet
features. It is found that 4-level wavelet decomposition is appropriate for feature vector
construction where all the recognition rates are the highest among almost all Daubechies
functions. In the closed-set FP recognition, the obtained recognition rates are below 90% due to
the imperfections in the FP images. To enhance the recognition rates further, a number of
preprocessing operations are used prior to wavelet transform and more than 96% recognition
rates are achieved in some Daubechies functions.
        The results have shown that the proposed WBSF method outperform the conventional
wavelet based recognition method. It seems that, the additionally selected bands provide extra
information and contribute in enhancing the recognition rates to attain 100% for D6, D8, ...,
D20 according to the test conditions considered in the work.
   The open-set FP verification mode is also presented for 290 trials from 29 persons. The
obtained verification rates, greater than 97%, using WBSF method are quite acceptable.

REFERENCES

[1] Mustafa D. Al-Hassani, “Identification Techniques using Speech Signals and Fingerprints”,
    Ph.D. Thesis, Department of Computer Science, Al-Nahrain University, Baghdad, Iraq,
    Sep. 2006.
[2] R. M. Mandi, S. S. Lokhande, "Rotation –Invariant Fingerprint Identification System",
    International Journal of Electronics Communication and Computer Technology (IJECCT),
    ISSN: 2249-7838, Vol. 2 Issue 4, July 2012.
[3] Rakesh Verma, Anuj Goel, "Wavelet Application in Fingerprint Recognition", International
    Journal of Soft Computing and Engineering (IJSCE), ISSN: 2231-2307, Vol. 1, Issue-4,
    Sep. 2011.
[4] Priti S. Sanjekar, Priyadarshan S. Dhabe, "Fingerprint Verification using Haar
    Wavelet",IEEE 2ndInternational Conference on Computer Engineering and Technology,
    Vol. 3, pp.361-365, 2010.


                                                                               322
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME

[5] D. Maltoni, D. Maio, A. K. Jain, S. Prabhakar, “Handbook of Fingerprint Recognition”,
    Springer Verlag, New York, 2003.
[6] Sagem Morpho Inc.,”Fingerprint Identification Technology in civil Applications”, 1145
    Broadway plaza Tacoma, Washington, 1998.
[7] M. Eriksson, “Biometrics Fingerprint based identity verification”, M. Sc. Thesis,
    Department of Computer Science, UMEÅUniversity, August 2001.
[8] A. Saeed, A. Tariq, U. Jawaid, "Automated System for Fingerprint Image Enhancement
    using Improved Segmentation and Gabor Wavelets", Department of Software Engineering,
    Fatima Jinnah University, Rawalpindi, Pakistan, 2008.
[9] F. A. Afsar, M. Arif and M. Hussain, “Fingerprint Identification and Verification System
    using Minutiae Matching”, Department of Computer & Information Sciences, Institute of
    Engineering & Applied Sciences, Islamabad, Pakistan, 2004.
[10] M. H. Ghassemian,“A Robust Structural Fingerprint Restoration”, Intelligent Signal
    Processing Research Center, Department of Electrical Engineering, Tarbiat Modares
    University, Iran, 1994.
[11] R. Setiono and H. Liu, ”Neural Network Feature Selector”, IEEE Transactions on
    Neural Network, Vol. 8, No.3, May 1997.
[12] A. M. Reza, “From Fourier Transform to Wavelet Transform: Basic Concepts”, White
    paper, Spire Lab, UWM, October, 1999.
[13] C. S. Burrus, R. A. Gopinath and H. Guo, “Introduction to Wavelets and Wavelet
    Transforms”, Prentice-Hall, Inc., U.S.A, 1998.
[14] Z. Ahmad, “A New Algorithm for Image Compression Based on Wavelet Transform”,
    M. Sc. Thesis, Electrical Engineering Department, University of Baghdad, Iraq, 1999.
[15] T. Chang and C.-C. Jay Kuo, “Texture Analysis and Classification with Tree-Structured
    Wavelet Transform”, IEEE Transactions on Image Processing, Vol. 2, No. 4, October 1993.
[16] M. P. Dale, M. A. Joshi, "Fingerprint Matching Using Transform Features", MES's
    College of Engineering, Pune, India, 2008.
[17] S. E. Umbaugh, “Computer Vision and Image Processing”, Prentice-Hall, Inc.,
    U.S.A., 1998.
[18] Rafael C. Gonzalez, Richard E. Woods, “Digital Image Processing”, Second Edition,
    Prentice-Hall, Inc., New Jersey, U.S.A., 2002.
[19] Mane Sameer S. and Dr. Gawade S.S., “Review on Vibration Analysis with Digital
    Image Processing”, International Journal of Advanced Research in Engineering &
    Technology (IJARET), Volume 4, Issue 3, 2013, pp. 62 - 67, ISSN Print: 0976-6480,
    ISSN Online: 0976-6499.
[20] Soukaena H. Hashem, Abeer T. Maolod and Anmar A. Mohammad, “Proposal To
    Enhance Fingerprint Recognition System”, International Journal of Computer Engineering
    & Technology (IJCET), Volume 4, Issue 3, 2013, pp. 10 - 22, ISSN Print: 0976 – 6367,
    ISSN Online: 0976 – 6375.
[21] Shekhar R Suralkar and Prof (Dr) Pradeep M. Patil, “Fingerprint Verification using
    Steerable Filters”, International Journal of Electronics and Communication Engineering &
    Technology (IJECET), Volume 4, Issue 2, 2013, pp. 264 - 268, ISSN Print: 0976- 6464,
    ISSN Online: 0976 –6472




                                           323

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:1
posted:6/27/2013
language:Latin
pages:16