6
Efficient Fingerprint Classification Using Singular Point Ali Ismail Awad and Kensuke Baba Graduate School of Information Science and Electrical Engineering, Kyushu University Library, Kyushu University 10-1, Hakozaki 6, Higashi-ku, Fukuoka, 812-8581, Japan {awad, baba}@soc.ait.kyushu-u.ac.jp ABSTRACT Keywords: Fingerprint Classification, Singular Point, Complex Fil- ter 1 INTRODUCTION Fingerprint as a kind of human biometrics on the finger tip is the dominant trait between many biometrics like iris, retina, and face. It is widely used for personal recognition in forensic and civilian applications because of its uniqueness, immutability, and low cost. Therefore, large volumes of fingerprints are collected and stored everyday by different means for a wide range of applications. In Automated Fingerprint Identification System (AFIS) with a large database, the input image should be matched against almost all fields inside the database to identify the most potential user identity. Recently, the AFIS systems that supporting instant identifica- tion or recognition are increasingly used, and the response time of these systems has become an important factor for measuring the system’s applicability. Although good performances have been measured for "one-to-one" matching, the time efficiency and the matching accuracy deteriorate seriously by extending the "one-to- one" matching to a "one-to-many" matching against a large database which may contain thousands of users [1]. In order to address the problem of reducing the system’s response time, fingerprint classi- fication has become an indispensable operation toward reducing the search time through large fingerprint databases. Fingerprint clas- sification refers to the problem of assigning fingerprint to one of several pre-specified classes. The classification process works on narrowing the search domain into smaller classes. By using finger- print classification, the input image will be matching only against one of the pre-specified classes. As the number of matching process get reduced, the system’s response time get shorter, accordingly. Fingerprint classification should be time efficient with a high clas- sification accuracy to avoid any bad impact on the system response time. Almost all of fingerprint classification approaches use one or more of the following features: singular points (core or delta points), orientation field, ridge flow, and the Gabor filter responses [2]. Us- ing the extracted features from singular points fall under one of strategies such as rule-based classifiers [3], [4], structural and sta- tistical approaches [5], and learning-based and multi classifier sys- tem [2], [6]. More recently, Neural Networks [1] and Wavelet- based classifiers [7] are also considered. The common fingerprint classes have been determined by Sir Henry [8] as Arch, Tended Arch, Left Loop, Right Lop, and Whorl. Fig.1 shows an example of fingerprint images in these classes. Most of the available classifica- tions methods are dominated by the nature of preprocessing steps that require a high processing time which slow down the overall system’s response time. In this context, some of the classification approaches archived a high classification accuracy, however, they have broken the time efficiency by consuming a longer processing time for the classification process. This paper presents a new fingerprint classification algorithm that uses the fingerprint texture (periodicity and directionality) prop- erty to create distinguished patterns for each fingerprint class. In order to emphases the texture property usage and avoid the intra- class problem of using full fingerprint image, a partitioning process is proposed to generate different patterns from each sub-image. The prototype of the presented algorithm [9] used the (x, y) lengths to divide fingerprint into equally four sub-images. Each class has four sub-images that constructing its distinguished patterns. The pro- 611 International Journal of Digital Information and Wireless Communications (IJDIWC) 1(3): 611-616 The Society of Digital Information and Wireless Communications, 2011 (ISSN 2225-658X) Singular point is one of the local fingerprint features, and it is used as a landmark due its scale and rotation immutability. Singular point charac- teristics have been widely used as a feature vector for many fingerprint classification approaches. This paper introduces a new fingerprint classi- fication method which utilizes a singular point as a reference point to part an input image. The key idea of the proposed classification method is di- viding fingerprint into small sub-images using singular point location, and then, creating distinguished patterns for each class using frequency domain representation for each sub-image. The performance evaluation has been conducted for the singular point detection method and the proposed clas- sification algorithm with different database. Both the processing time and the classification accuracy are considered as key issues of the classification approach. The experimental work shows that the achieved classification ac- curacy over FVC2002 database subsets is up to 83.7% with considerable processing time and robustness to scale, shift, and rotation.

EFFICIENT FINGERPRINT CLASSIFICATION USING SINGULAR POINT

  • Upload
    ijdiwc

  • View
    29

  • Download
    2

Embed Size (px)

DESCRIPTION

Singular point is one of the local fingerprint features, and it is used as alandmark due its scale and rotation immutability. Singular point characteristicshave been widely used as a feature vector for many fingerprintclassification approaches. This paper introduces a new fingerprint classificationmethod which utilizes a singular point as a reference point to partan input image. The key idea of the proposed classification method is dividingfingerprint into small sub-images using singular point location, andthen, creating distinguished patterns for each class using frequency domainrepresentation for each sub-image. The performance evaluation has beenconducted for the singular point detection method and the proposed classificationalgorithm with different database. Both the processing time andthe classification accuracy are considered as key issues of the classificationapproach. The experimental work shows that the achieved classification accuracyover FVC2002 database subsets is up to 83:7% with considerableprocessing time and robustness to scale, shift, and rotation.

Citation preview

Page 1: EFFICIENT FINGERPRINT CLASSIFICATION USING SINGULAR POINT

Efficient Fingerprint Classification Using Singular Point

Ali Ismail Awad† and Kensuke Baba‡

† Graduate School of Information Science and Electrical Engineering, Kyushu University‡ Library, Kyushu University

10-1, Hakozaki 6, Higashi-ku, Fukuoka, 812-8581, Japan

{awad, baba}@soc.ait.kyushu-u.ac.jp

ABSTRACT

Keywords: Fingerprint Classification, Singular Point, Complex Fil-

ter

1 INTRODUCTIONFingerprint as a kind of human biometrics on the finger tip is the

dominant trait between many biometrics like iris, retina, and face.

It is widely used for personal recognition in forensic and civilian

applications because of its uniqueness, immutability, and low cost.

Therefore, large volumes of fingerprints are collected and stored

everyday by different means for a wide range of applications. In

Automated Fingerprint Identification System (AFIS) with a large

database, the input image should be matched against almost all

fields inside the database to identify the most potential user identity.

Recently, the AFIS systems that supporting instant identifica-

tion or recognition are increasingly used, and the response time

of these systems has become an important factor for measuring

the system’s applicability. Although good performances have been

measured for "one-to-one" matching, the time efficiency and the

matching accuracy deteriorate seriously by extending the "one-to-

one" matching to a "one-to-many" matching against a large database

which may contain thousands of users [1]. In order to address the

problem of reducing the system’s response time, fingerprint classi-

fication has become an indispensable operation toward reducing the

search time through large fingerprint databases. Fingerprint clas-

sification refers to the problem of assigning fingerprint to one of

several pre-specified classes. The classification process works on

narrowing the search domain into smaller classes. By using finger-

print classification, the input image will be matching only against

one of the pre-specified classes. As the number of matching process

get reduced, the system’s response time get shorter, accordingly.

Fingerprint classification should be time efficient with a high clas-

sification accuracy to avoid any bad impact on the system response

time.

Almost all of fingerprint classification approaches use one or

more of the following features: singular points (core or delta points),

orientation field, ridge flow, and the Gabor filter responses [2]. Us-

ing the extracted features from singular points fall under one of

strategies such as rule-based classifiers [3], [4], structural and sta-

tistical approaches [5], and learning-based and multi classifier sys-

tem [2], [6]. More recently, Neural Networks [1] and Wavelet-

based classifiers [7] are also considered. The common fingerprint

classes have been determined by Sir Henry [8] as Arch, Tended

Arch, Left Loop, Right Lop, and Whorl. Fig.1 shows an example of

fingerprint images in these classes. Most of the available classifica-

tions methods are dominated by the nature of preprocessing steps

that require a high processing time which slow down the overall

system’s response time. In this context, some of the classification

approaches archived a high classification accuracy, however, they

have broken the time efficiency by consuming a longer processing

time for the classification process.

This paper presents a new fingerprint classification algorithm

that uses the fingerprint texture (periodicity and directionality) prop-

erty to create distinguished patterns for each fingerprint class. In

order to emphases the texture property usage and avoid the intra-

class problem of using full fingerprint image, a partitioning process

is proposed to generate different patterns from each sub-image. The

prototype of the presented algorithm [9] used the (x, y) lengths to

divide fingerprint into equally four sub-images. Each class has four

sub-images that constructing its distinguished patterns. The pro-

611

International Journal of Digital Information and Wireless Communications (IJDIWC) 1(3): 611-616The Society of Digital Information and Wireless Communications, 2011 (ISSN 2225-658X)

Singular point is one of the local fingerprint features, and it is used as alandmark due its scale and rotation immutability. Singular point charac-teristics have been widely used as a feature vector for many fingerprintclassification approaches. This paper introduces a new fingerprint classi-fication method which utilizes a singular point as a reference point to partan input image. The key idea of the proposed classification method is di-viding fingerprint into small sub-images using singular point location, andthen, creating distinguished patterns for each class using frequency domainrepresentation for each sub-image. The performance evaluation has beenconducted for the singular point detection method and the proposed clas-sification algorithm with different database. Both the processing time andthe classification accuracy are considered as key issues of the classificationapproach. The experimental work shows that the achieved classification ac-curacy over FVC2002 database subsets is up to 83.7% with considerableprocessing time and robustness to scale, shift, and rotation.

Page 2: EFFICIENT FINGERPRINT CLASSIFICATION USING SINGULAR POINT

Figure 1. The common five fingerprint classes with singular points representation.

posed algorithm invests the scale and rotation invariant property of

the singular point by considering it as a reference point for image

division process.

In the proposed algorithm, we give an attention to both the time

efficiency and the classification accuracy. The algorithm is opti-

mized for time efficiency by selecting an efficient method for sin-

gular point detection [10]. With respect to the classification accu-

racy, selecting the singular point as a reference for the image divid-

ing process makes the algorithm more robust for image scale and

rotation. The experimental work shows that the proposed method

competes well with available classification approaches in terms of

the classification accuracy and the processing time.

The reminder part of this paper is organized as follows: Sec-

tion 2 describes the proposed classification algorithm including the

singular point detection. The experimental work is described in

Section 3. Finally, the conclusions are reported in Section 4.

2 ALGORITHMThe main idea of our classification method is dividing an input im-

age based on the position of a singular point. Each sub-image is

processed individually using 2D-FFT to produce the pattern shape.

The frequency patterns are scale and rotation invariant since it is

limited to the texture orientation inside each block. Therefore, us-

ing singular point location will guarantee that the process of creat-

ing class patterns will be also robust for scale and rotation. The

frequency patterns are processed by Gabor filter [11] to extract

their orientations. Finally, classification process is carried out by

matching the pattern of the input fingerprint with the patterns of

the five classes. The proposed classification method is flexible, and

the number of output classes can be easily extended by increasing

the number of the standard classes.

2.1 Singular Point DetectionThe singular points (core and delta) shown in Fig.1 are the most

important global characteristics of fingerprint images. A singular

point area is defined as a region where the ridge curvature is higher

than normal and where the direction of the ridge changes rapidly.

A core point is defined as the top-most point of the innermost curv-

ing ridge, and a delta point is the center of triangular regions where

three different direction flows meet [2]. As the algorithm to de-

tect a singular point in an input image, we use the existing method

with complex filters [12]. The Poincaré index method [13] is the

most common way for singular point detection. One of the differ-

ences between the two methods is that the complex filter method

searches the two kinds of singular point separately. The classifica-

tion method in this paper utilizes the location (position) of a single

core point, hence the required output of singular point detection is

at least the position of one core point. By the requirement, if we

use the complex filter method for singular point detection in our

method, we can skip almost half of the processes of the method for

searching delta points. We actually compared the processing time

and the accuracy of the two methods on the assumption of usage

for our classification method, and we found that the complex filter

method is suitable in both senses compared with the Poincaré index

method [10].

The complex filter method detects symmetric parts in the com-

plex orientation field by applying two kinds of complex filters for

core and delta points. The complex orientation field image is ob-

tained from the input image by

z(x, y) = (fx + ify)2, (1)

where i is the imaginary unit and fx and fy are the derivatives of the

input image in the x- and y-directions, respectively. The complex

612

International Journal of Digital Information and Wireless Communications (IJDIWC) 1(3): 611-616The Society of Digital Information and Wireless Communications, 2011 (ISSN 2225-658X)

Page 3: EFFICIENT FINGERPRINT CLASSIFICATION USING SINGULAR POINT

filters for core points is

Fc = (x+ iy)mg(x, y), (2)

where m is the filter order and g is the Gaussian such that

g(x, y) = exp

(−x

2 + y2

2σ2

)(3)

with the variance σ. The convolution of the complex orientation

field image with the complex filter is computed, and then a point

with high filter response is considered as a core point.

In this method, the computation of the convolution takes an

O(n2 logn) time for the size n2 of the input image, and the other

process are constant or linear to n2. Therefore the time complexity

of the method is O(n2 logn).

2.2 Pattern ExtractionPrior to the pattern extraction, an accurate image partitioning has

been done based on the singular point location. To ensure the cor-

rectness of image partitioning, image padding has been also imple-

mented on all images to make sure that all sub-images are individ-

ual square blocks. The pattern of an image is constructed from the

frequency components of the four sub-images, that is, the pattern

is a 4-tuple of images. The pattern for each category of fingerprint

classification is extracted from classified images in advance, and it

is called a standard pattern.

The first step toward pattern extraction is frequency domain

representation. The two dimensional Fast Fourier Transform (2D-

FFT) [14] has been applied individually on each sub-image. The

resulted patterns are defined by both the direction and the shape.

Fig. 2 shows the outputs each step in pattern extraction process,

and moreover. The second step of pattern extraction is making an

adaptive threshold to the FFT output components. This step aims

to primary segment the foreground components in each sub-image.

By a simple visual inspection on Fig. 2(c) and Fig. 3, one can real-

ize how much the 2D-FFT outputs of the sub-images are different

from the full image. Fig. 2(d) shows an example of the final ex-

tracted patterns after we perform an adaptive threshold process.

2.3 Pattern MatchingIn order to reduce the processing time of the matching, the orienta-

tion of a pattern image has been considered as the first step of the

matching. Due to the texture properties of the extracted patterns,

the Gabor filter [11] has been applied to separate a pattern image

into lines like parallel ridges and valleys, and then the direction is

extracted using the Sobel filter [14].

A general form of 2D Gabor filter can be split into odd and

even symmetric components. In our method, the odd-symmetric

components will be considered. The odd-symmetric components

(a) (b)

(c) (d)

Figure 2. The output of each algorithm’s step: Singular point detection (a),dividing image into four sub-images (b), final extracted patterns (c), andfrequency representation of each sub-image (d).

(a) (b)

Figure 3. Frequency pattern of using full fingerprint image.

of the Gabor filter [15] is

g(x, y) = exp

{−1

2

(x2φσ2x

+y2φσ2y

)}sin

{2πxφT

}, (4)

where φ is the filter orientation, T is the filter wavelength, and

xφ = x cosφ+ y sinφ, yφ = −x cosφ+ y sinφ (5)

One of the Gabor filter properties is the performing of low pass

filtering along the orientation φ and band pass filtering orthogonal

to that orientation. We got the result in Fig. 4(a) by configuring the

Gabor filter as T = 10 pixels, φ = 45◦, and σ = 1 for both x and

y coordinates. The final result has been produced by implementing

the Gabor filter using 2D convolution on the extracted patterns. The

result of applying the Sobel filter is of the size 16 × 16 is shown

in Fig.4 (b). Fig.5 shows examples of the standard patterns. The

orientations (θ1, θ2, θ3, θ4) of the patterns are (130◦, 45◦, 120◦,

90◦) for Arch, (135◦, 48◦, 135◦, 0◦) for Left Loop, (135◦, 48◦,

0◦, 0◦) for Right Loop, (135◦, 45◦, 45◦, 0◦) for Whorl, and (135◦,

45◦, 45◦, 135◦) for Twin Loop, where the θ subscript expresses the

sub-image number in the dashed circle in Fig.5.

613

International Journal of Digital Information and Wireless Communications (IJDIWC) 1(3): 611-616The Society of Digital Information and Wireless Communications, 2011 (ISSN 2225-658X)

Page 4: EFFICIENT FINGERPRINT CLASSIFICATION USING SINGULAR POINT

(a) (b)

Figure 4. Pattern transformation into ridge like shape using Gabor filter:the output of Gabor filter (a) and the extracted pattern orientation (b).

(a) (b)

(c) (d)

(e)

Figure 5. The patterns of the standard classes: Arch (a), Left Loop (b),Right Loop (c) Whorl (d), and Twin Loop (e).

In order to increase the time efficiency, the pattern matching

has been conducted in two consecutive phases. The first phase uses

the pattern orientation, the orientation of each sub-image in the in-

put image is matched with its corresponding one of the standard

class. The two orientations are considered as matched if the differ-

ence is less than the tolerance 5◦. This phase is very time efficient

because it numerically processes the pattern orientations. In the

case that no orientation is matched, the second phase is applied.

The second phase is a correlation matching for each sub-image.

The total correlation between the input image and the standard class

is the summation of the four patterns correlations.

3 EXPERIMENTAL RESULTSThe application of singular points in fingerprint classification has

two main steps, singular point detection and fingerprint classifica-

tion. During our experiments, both steps have been evaluated in

terms of the processing time and the accuracy, respectively.

The samples for the experiments are three subsets of Finger-

print Verification Competition 2002 [16] DB1_B, DB2_B, and

DB3_B which have been captured by an optical sensor "TouchView

II" by Identix, "FX2000" by Biometrika, and a capacitive sensor

"100 SC" by Precise Biometrics, respectively. The number of the

fingerprint images in each subset is 80.

3.1 Processing TimeThe processing time has been measured in two phases: the singular

point detection phase, and the overall processing time for the classi-

fication method. The experiments have been operated with IntelTM

Core 2 DuoTM

Processor (T9300, 2.5 GHz, 6 MB L2 cash), 3 GB

of RAM, Windows XP R© Pro 32 bit, and MATLAB R© R2009b ver-

sion.

Table 1 summarizes the processing times of singular point de-

tection used in the proposed method with the three databases. Ta-

ble 1 points out that the singular point detection method consumes

very short processing time compared to the Poincaré index method

[10]. It is worth noticing that the processing time is also very small

against to the method in [17] which takes 10 seconds for processing

a single fingerprint image.

Table 1. The processing time of singular point detection by the proposedmethod (second).

Min Max Average

DB1_B(80) 0.11 0.14 0.12

DB2_B(80) 0.12 0.16 0.13

DB3_B(80) 0.08 0.09 0.08

In the context of the total processing time, Table 2 represents

the experimental result about the processing time of the proposed

method. The processing time has been measured for all concurrent

algorithm’s steps including the singular point detection time. By

Table 2, the total processing time is very short compared to [1], and

[18], which consume 1.6 and 5 seconds, respectively.

Table 2. The processing time of fingerprint classification by the proposedmethod (second).

SP Image Patterns Patterns Total

detect divide extract match time

Time/Image 0.11 0.02 0.08 0.07 0.28

614

International Journal of Digital Information and Wireless Communications (IJDIWC) 1(3): 611-616The Society of Digital Information and Wireless Communications, 2011 (ISSN 2225-658X)

Page 5: EFFICIENT FINGERPRINT CLASSIFICATION USING SINGULAR POINT

3.2 AccuracyThe accuracy of singular point detection is measured by the num-

ber of the correctly detected images in the total images. Table 3

shows the results of the accuracy of singular point detection in the

proposed method. The correct position of a core point was checked

for each image in advance by a visual inspection. In the “Rejected”

images, any core point couldn’t be detected even by a visual in-

spection. The accuracy with the three database are 91.25 %, 95.0

%, and 90.0 %, respectively.

Table 3. The accuracy of singular point detection by the proposed method.

(240) Fingerprint ImagesDatabase Detected Incorrect Rejected Accuracy

DB1_B(80) 73 7 0 91.25 %

DB2_B(80) 76 4 0 95 %

DB3_B(80) 63 7 10 90 %

The accuracy of the classification has been evaluated with the

same databases. Prior to the accuracy of the total classification,

we first examine the accuracy of the process after the singular point

detection, that is, the rejected images in the singular point detection

process are not considered. Fig. 6 shows some samples of rejected

image in the both cases. Tables 4, 5, and 6 show the results of

classification by the proposed method with DB1_B, DB2_B, and

DB3_B, respectively. The true class of each image was checked in

advance by a visual inspection. In each database, respectively, 10,

10, and 22 images have been excluded as ambiguous images (that

is, the images couldn’t be classified even by a visual inspection).

Therefore, the number of sample images are 70, 70, and 58. The

accuracy of the proposed method is 91.4 %, 88.5 %, and 82.7 %

with 12.5 %, 12.5 %, and 27.7 % rejection rates, respectively.

Table 4. The accuracy of fingerprint classification by the proposed methodwith DB1_B.

Assigned ClassesTrue Classes Arch R. L. L. L. W. T. L.

Arch (0) 0 0 0 0 0

Right Loop (34) 1 30 2 1 0

Left Loop (20) 0 1 19 0 0

Whorl (8) 0 0 1 7 0

Twin Loops (8) 0 0 0 0 8

The total classification accuracy of the proposed method is also

considered. Let C be the number of the correctly classified images,

I the number of the incorrectly classified images, and S the num-

ber of the images with incorrect singular points. Then, the total

Table 5. The accuracy of fingerprint classification by the proposed methodwith DB2_B.

Assigned ClassesTrue Classes Arch R. L. L. L. W. T. L.

Arch (0) 0 0 0 0 0

Right Loop (26) 0 23 1 2 0

Left Loop (22) 1 0 18 1 2

Whorl (16) 0 0 1 15 0

Twin Loops (6) 0 0 0 0 6

Table 6. The accuracy of fingerprint classification by the proposed methodwith DB3_B.

Assigned ClassesTrue Classes Arch R. L. L. L. W. T. L.

Arch (17) 13 0 1 3 0

Right Loop (8) 0 6 0 2 0

Left Loop (1) 0 0 0 0 0

Whorl (32) 1 0 1 29 1

Twin Loops (0) 0 0 0 0 0

classification accuracy At is measured as

At =C

C + I + S(6)

Table 7 shows the total classification accuracies for the three

database subsets. The maximum achieved accuracy is 83.7% com-

pared to 88.3% with using singular point only [19] which is in an

acceptable range. That table expresses the effect of the accuracy

of the singular point detection method on the total classification

accuracy. The classification accuracies of DB1_B and DB3_B are

highly degraded because they have more incorrect detected singu-

lar points. Although, the classification accuracy of DB2_B has lit-

tle degradation compared to the other databases due to the small

amount of incorrect singular points.

Table 7. The total accuracy of fingerprint classification by the proposedmethod.

Database AccuracyDB1_B 83.1 %

DB2_B 83.7 %

DB3_B 73.8 %

4 CONCLUSIONSThis paper presented a new fingerprint classification algorithm that

generates a distinguished classification patterns from the ridge fre-

quency and the ridge orientation. The main idea of the proposed

method is dividing an image into four sub-images by a singular

615

International Journal of Digital Information and Wireless Communications (IJDIWC) 1(3): 611-616The Society of Digital Information and Wireless Communications, 2011 (ISSN 2225-658X)

Page 6: EFFICIENT FINGERPRINT CLASSIFICATION USING SINGULAR POINT

Figure 6. Examples of excluded fingerprint images: rejected images (top) and ambiguous images (down).

point location. The partitioning process introduces some extra pro-

cessing time, however the extracted patterns from four sub images

are completely different from using a full image patterns with intra-

class avoidance. We also examined the processing time and the

accuracy of the proposed method by experiments with standard

databases of fingerprint images. As the result, the proposed method

processed in a very short time compared with other existing meth-

ods with an acceptable accuracy.

5 REFERENCES1. Liu, M.: Fingerprint classification based on Adaboost learning from

singularity features. Pattern Recognition 43(3), 1062–1070 (2010)

2. Maltoni, D., Maio, D., Jain, A.K., Prabhakar, S.: Handbook of Fin-gerprint Recognition. Springer-Verlag, second edn. (2009)

3. Jain, A.K., Minut, S.: Hierarchical kernel fitting for fingerprint classi-fication and alignment. Pattern Recognition, International Conferenceon 2, 20469 (2002)

4. Cappelli, R., Maio, D., Maltoni, D.: A Multi-Classifier approach tofingerprint classification. Pattern Analysis & Applications, SpringerVerlag London 5(2), 136–144 (2002)

5. Maio, D., Maltoni, D.: A structural approach to fingerprint classifica-tion. In: Proceedings of the 13th International Conference on PatternRecognition. Vol. 3, pp. 578–585. IEEE, Vienna, Austria (1996)

6. Cappelli, R., Maio, D., Maltoni, D., Nanni, L.: A two-stage finger-print classification system. In: Proceedings of the 2003 ACM SIGMMworkshop on Biometrics methods and applications (WBMA ’03).Vol. 1, pp. 95–99. ACM, Berkley, California, USA (2003)

7. Wang, W., Li, J., Chen, W.: Fingerprint classification using improveddirectional field and fuzzy wavelet neural networks. In: Proceedingsof the IEEE 6th World Congress on Intelligent Control and Automa-tion. Vol. 2, pp. 9961–9964. IEEE, Dalian, China (2006)

8. Henry, E.R.: Classification and uses of Fingerprints. Routledge &

Sons, London (1900)

9. Awad, A.I., Baba, K.: Toward an efficient fingerprint classification.In: Albert, M. (ed.) Biometrics - Unique and Diverse Applications inNature, Science, and Technology. InTech (2011)

10. Awad, A.I., Baba, K.: Fingerprint singularity detection: A com-parative study. In: Zain, J.M., bt Wan Mohd, W.M., El-Qawasmeh,E. (eds.) Software Engineering and Computer Systems. Communica-tions in Computer and Information Science, Vol. 179, pp. 122–132.Springer (2011)

11. Gabor, D.J.: Theory of communication. IEE 93(26), 429–457 (1946)

12. Nilsson, K., Bigun, J.: Localization of corresponding points in finger-prints by complex filtering. Pattern Recognition Letters 24(13), 2135–2144 (2003)

13. Kawagoe, M., Tojo, A.: Fingerprint pattern classification. PatternRecognition 17(3), 295–303 (1984)

14. Gonzalez, R.C., Woods, R.E., Eddins, S.L.: Digital Image ProcessingUsing Matlab. Prentice Hall, first edn. (ISBN: 0130085197, 2003)

15. Yang, J., Liu, L., Jiang, T., Fan, Y.: A modified gabor filter designmethod for fingerprint image enhancement. Pattern Recognition Let-ters 24(12), 1805–1817 (2003)

16. Maio, D., Maltoni, D., Cappelli, R., Wayman, J.L., Jain, A.K.:FVC2002: second fingerprint verification competition. In: Proceed-ings of the 16th International Conference of Pattern Recognition(ICPR). pp. 811–814 (2002)

17. Magalhaes, F., Oliveira, H., Campilho, A.: A new method for thedetection of singular points in fingerprint images. In: IEEE Workshopon Applications of Computer Vision (WACV). pp. 1–6 (2009)

18. Rao, K., Balck, K.: Type classification of fingerprints: A syntacticapproach. IEEE Transactions on Pattern Analysis and Machine Intel-ligence PAMI-2(3), 223–231 (1980)

19. Li, J., Yau, W., Wang, H.: Combining singular points and orientationimage information for fingerprint classification. Pattern Recognition41(1), 353–366 (2008)

616

International Journal of Digital Information and Wireless Communications (IJDIWC) 1(3): 611-616The Society of Digital Information and Wireless Communications, 2011 (ISSN 2225-658X)