Вы находитесь на странице: 1из 4

2009 International Conference on Computational Intelligence and Natural Computing

Fingerprint Segmentation using Cellular Neural


Network
Jiayin Kang1

Wenjuan Zhang2

School of Electronics Engineering,


Huaihai Institute of Technology,
Lianyungang, China
jiayinkang@gmail.com

School of Computer Engineering,


Huaihai Institute of Technology,
Lianyungang, China
Zhangwenjuan200@163.com

groups, i.e., the gray-level values below the selected threshold


are usually classified as background (become white, -1), while
the values above this threshold are classified as object (become
black, +1).

Abstract: Fingerprint segmentation is a crucial step in fingerprint


recognition system, and determines the results of fingerprint analysis
and recognition. In this paper, an approach for fingerprint
segmentation based on cellular neural network (CNN) is presented.
The proposed algorithm is realized by threshold CNN (T-CNN), of
which threshold is obtained via fuzzy set theory. Experimental results
show that proposed method can extract the fingerprint effectively and
efficiently from original image.

However, in real life situations great deals of images are


ambiguous and usually have indistinguishable histogram.
Therefore, it is difficult to determine the threshold of the TCNN.

Keywords: fingerprint; segmentation; cellular neural network;


fuzzy set; threshold

I.

In this paper, we presented an approach for fingerprint


segmentation using CNN-based thresholding, in which the
threshold of CNN is calculated via fuzzy set theory.
Experimental results show that proposed method can extract
the fingerprint effectively and efficiently from original image.

INTRODUCTION

Fingerprint segmentation is an important issue in


fingerprint recognition system. A fingerprint image usually has
to be segmented to remove uninterested regions before some
other steps such as enhancement and minutiae detection so that
the image processing will consume less CPU time. A
fingerprint image generally consists of different regions: nonridge regions, high quality ridge regions, and low quality ridge
regions. Fingerprint segmentation is usually to identify nonridge regions and unrecoverable low quality ridge regions and
exclude them as background [1]. Most segmentation methods
are block-wised ones which divide the fingerprint image into
un-overlapped blocks and decide on the type (background and
foreground) of each block. And some other methods are pixelwised ones which determine the type of each pixel. Fingerprint
segmentation typically computes the feature (or feature vector)
of each element, block or pixel, and then determine the
elements type based on the feature (vector). The features used
in fingerprint segmentation mainly include statistical features
of pixel intensity, directional image and ridge projection signal
et al.

A.

Cellular Neural Network


The standard MN CNN architecture is composed of cells
C (i, j )' s . The dynamic of each cell is given via the equation
as follows [2]:

xi , j = xi , j +

ak ,l yi + k , j +l +

k ,lSi , j ( r )

bk ,l ui + k , j +l + zi, j

k ,lSi , j ( r )

(1)

= xi , j + ak ,l yi + k , j +l
k = r l = r

+ bk ,l ui + k , j +l + zi , j

Cellular neural network[2], locally connected (take


neighboring spatial information into account), first introduced
as an implemental alternative to full-connected Hopfield neural
network, has been widely studied for image processing, robotic
and biological vision, and higher brain functions [3].

k = r l = r

i = 1, 2, ", M ; j = 1, 2, ", N

xi , j , y i , j , u i , j and z i , j represent state, output, input,


and threshold respectively; Si , j ( r ) is the sphere of influence
Where

In a Chuas exposition (see [3] or [4]), many important and


interesting CNNs are described. One of them is the threshold
CNN (T-CNN), which can extract the object from the
background in gray-level image. The key point of T-CNN is to
determine the threshold value which assign pixels into two
978-0-7695-3645-3/09 $25.00 2009 IEEE
DOI 10.1109/CINC.2009.246

THRESHOLDING CNN

II.

with radius r ;

11

'
'
a k ,l s and bk ,l s are the elements of the A-

III.

template and the B-template respectively. The output y i , j is


the piece-wise linear function given by

yi , j =

OBTAINING THRESHOLD IN T-CNN VIA FUZZY SET


THRORY

A. Gray-Scale Histogram

1
(| xi , j + 1| | xi , j 1|)
2

(2)
Image histogram is one of the most important statistical
characteristics of image, and means the statistical relationship
between each gray scale and its corresponding occurrence
frequency in the image [5]. The histogram of the image is a
discrete function defined as follow:

B. Standard Threshold CNN (T-CNN)


The template of the standard threshold CNN has the form
[4]:

H ( sk ) = nk ,

k = 0,1,", L 1

(4)

sk is gray value of the k th gray scale; nk is the


number of the pixels with gray value sk , i.e., the occurrence
frequency of the k th gray scale; k is the gray scale and L is
Where

Where

1 < z* < 1

(3)

Generally, dividing each gray scales occurrence frequency


by the total number of pixels n , normalized histogram can be
obtained in the form as follows:

I. Global Task
1. Given:

Static gray-scale image P and threshold

*.

2. Input:

U (t ) = 0 .
3. Initial State:

U (t ) = Arbitrary

or

default

the maximum gray scale.

P ( sk ) =

to

nk
n

k = 0,1, ", L 1

From Equation (5), it is indicate that

X ( 0) = P .

(5)

P( sk ) is the

occurrence probability of k th gray scale in the image.


Therefore, normalized histogram provides the distribution of
the gray scale in the image, i.e., provides a whole description of
the gray scale in the image, as shown in figure 1.

Y (t ) Y () = binary image when all


*
pixels with gray-scale intensity Pi , j > z become black.
4. Output:

II. Local Rules

xi , j (0)
1.

xi , j (0) < z *

y i , j ( )

White,

independent

of

Black,

independent

of

neighbors.
2.

xi , j (0) > z *

neighbors.
3.

xi , j (0) = z *

Figure 1.

z * , assuming zero noise.

From the form of the T-CNN as shown in equation (3), it is


clear that successful image thresholding mainly depends on
rational selection of the threshold z * .

Histogram of the gray-scale image

B. Obtaining the Threshold in T-CNN via Fuzzy Set Theory


Roughly speaking, a digital image can be divided into two
groups: one stands for the background, and another stands for
the object. Correspondingly, there are two peaks and one valley
in the image histogram. as shown in Fig. 1, T1 and T2 are two
gray scale (horizontal coordinate of histogram) corresponding
to two peaks in the gray-scale image histogram respectively. T
is gray scale corresponding to valley. Actually, T1 is the
backgrounds center, and T2 is the foregrounds center.
Consequently, T is the threshold by which assign pixels into
two groups, i.e., the gray-level values above T are usually

Generally, the gray-level histogram of an image may have


one or more peak values. By choosing one or more thresholds
of gray level, the object regions can be classified for further
processing. If the histogram has obvious peak values, we can
get the thresholds easily. Nevertheless, because of noise in
practice, we could not always make a good choice of the
threshold. The threshold could easily be influenced by the
noise.

12

classified as background (become white, 255), while the values


below this threshold are classified as object (become black, 0).

fuzzy set theory described in section III.B, we performed the


fingerprint segmentation experiments.

Therefore, how to obtain the threshold T is the key issue in


image thresholding. For the purpose of computation of
threshold T, we adopted the fuzzy thresholding algorithm
described as follows:

300 300 . Correspondingly, fingerprint segmentation results

In the notation of fuzzy set, the image set


size M N can be written as

Fig. 2(a) show the fingerprint images with the size of

using Otsus[7], Kapurs[8] and proposed method are displayed


in Figs. 2(b)-(d) respectively.

I = f ( x, y ) of

I = {( f ( x, y ), I ( f ( x, y )))}

As shown in Fig. 2, it is clear that our proposed algorithm


performs more visually significant than other two methods do.
Fig. 3 (a) display another fingerprint image with the size of

(6)

300 300 . Correspondingly, fingerprint segmentation results

using three methods, Otsus, Kapurs and proposed method, are


displayed in Figs. 3(b)-(d) respectively. The same conclusion
can be reached, i.e., fingerprint segmentation result using our
proposed method is visually clear while other two methods
based are illegible.

The membership function I ( f ( x, y )) can be viewed as


a characteristic function that represents the fuzziness of a
( x, y ) pixel in I .
1

1+ | f ( x, y ) (t ) | C

0
I ( f ( x, y )) =
1

1+ | f ( x, y ) 1 (t ) | C

if f ( x, y ) t

(7)
if f ( x, y ) > t

Where the average gray levels 0 (t ) and 1 (t ) can be


regarded as the target values of the background and the object
for a given threshold value t ; and C is a constant value such
that

0.5 I ( f ( x, y )) 1 .
t

i =0

i =0

0 (t ) = i h(i ) / h(i )
1 (t ) =

L 1

L 1

i =t +1

i =t +1

i h(i) / h(i)

(a)

(b)

(c)

(d)

(8)

(9)

The measure of fuzziness that was used in this work was


the entropy E ( I ) by using the Shannons function

S ( ( xi )) [6]. The optimal threshold can be determined by


minimizing the measure of fuzziness E ( X ) .
S ( I (i )) = I (i ) ln[ I (i )] [1 I (i )]ln[1 I (i )]

E(I ) =

L 1
1
S ( I (i))h(i)
MN ln 2 i = 0

t = arg min E ( X )
tGL

(10)

Figure 2. Fingerprint image segmentation . (a) original fingerprint image. (b)


fingerprint segmentation result using Otsus method. (c) fingerprint
segmentation result using Kapurs method. (d) fingerprint segmentation result
using proposed method.

(11)
(12)

As for CNN, it is note that the white pixel gray value equals
to -1 and the black pixel gray value equals to 1, that is,
1 < z * < 1 . Therefore, obtain the threshold z * in T-CNN as:

z * = 2 (t / 255 0.5)
IV.

(13)

FINGERPRINT SEGMENTATION VIA CNN

(a)

Based on both the theory of threshold CNN described in


section II.B and the proposed threshold obtaining algorithm via

13

(b)

ACKNOWLEDGMENT
The authors would like to address appreciation to
anonymous reviewers for their valuable and helpful comments.
References
[1]

[2]

(c)

[3]

(d)

[4]

Figure 3. Fingerprint image segmentation . (a) original fingerprint image. (b)


fingerprint segmentation result using Otsus method. (c) fingerprint
segmentation result using Kapurs method. (d) fingerprint segmentation result
using proposed method.

[5]
[6]

V.

CONCLUSION

In this paper, an approach for fingerprint segmentation


based on cellular neural network associated with fuzzy set
theory was proposed. The proposed algorithm is realized by
threshold CNN, of which threshold was obtained via fuzzy
thresholding algorithm automatically. Experimental results
show that proposed method performs more visually effective
during extracting the fingerprint from original image than other
two methods do.

[7]
[8]

14

J. Yin, X. Yang, G. Zhang, C. Hun, Two steps for fingerprint


segmentation, Image and Vision Computing, Vol. 25, pp. 13911403,
2007.
L. O. Chua, L. Yang. Cellular neural networks: theory and
applications, IEEE trans. circuits and system, Vol. 35, pp. 1257-1290,
1988.
L. O. Chua. CNN: a vision of complex, Int. J. bifurcation and chaos,
Vol. 7, No. 10, pp. 2219-2425, 1997.
L. O. Chua, T. Roska. Cellular neural networks and visual computing.
Cambridge Press, London, 2002.
M. Yao. Digital image processing ( In Chinese ). Mechanical Industry
Publishing House, Beijing, 2006.
Ursula Gonzales-Barron, Francis Butler. A comparison of seven
thresholding techniques with the k-means clustering algorithm for
measurement of bread-crumb features by digital image analysis,
Journal of Food Engineering, Vol. 74, pp. 268278, 2006.
N. Otsu. A threshold selection method for grey level histograms, IEEE
Trans. Syst. Man Cybern, SMC-9 (1) , pp. 6266, 1979.
E.J.N. Kapur, P.K. Sahoo, A.K.C. Wong. A new method for greylevel
picture thresholding using the entropy of the histogram, Comput. Vis.
Graph. Image Process, Vol. 29 , pp. 273285, 1985.

Вам также может понравиться