Вы находитесь на странице: 1из 4

A New Approach of Finger-Print Recognition based on Neural Network

Behnaz Saropourian
Dept.Computer Engineering,Islamic Azad University,Kermanshah-Branch
Islamic Azad University
Kermanshah, Iran
e-mail: behnazsaropourian@gmail.com

Abstract—in this issue opposite of other custom and classic


patterns of the finger-print recognition that they have working on II. FINGER-PRINT DETAILS
the single dots existing on the finger-print liked ridge-ending and
bifurcation. This new observation is about finger-print The finger-print will have recognized pattern of projections and
recognition that uses state of the single steak in the finger-print depressions on surface of finger. [4]
image, because patterns of the veins in the finger-print without The fixing similar of two samples of finger-print are difficultly
notice of the finger-print about person, and can be used for based on images. This method is requiring a long process. In the
identified person. This algorithm can work very fine on the binary other hand recognized of finger-print must be based on basic
images and also gray scanned photo. specifications that named minutae details.
The minutae details are defining liked that intersection,
Keywords—Biometric, Neural Network, Multilayer Precpetron, ledge-ending, dots, island, stiff, bifurcation, bridge, cerebral
Error Propagation
ledge. These details are show as follows:

I. INTRODUCTION
In the early years, many of the searching for individual based
on individual traits. Biometric systems fundamentally is a
pattern recognition systems, that person can recognized that
based on special traits physiology or behavioral.
Biometric systems defined by IEEE in the form of below:
Biometric systems are automatic systems to recognized person
based on physical traits and behavioral traits. [1] Physical
biometric working with based on information from part of body Figure.1 Finger-Print details
liked finger-print, face, iris and retina. But behavioral biometric The fixing place and kind finger-print details are made
available a correct and pressed representation.
working with based on information from individual activity
In this cause, we used two kinds of ridge of special points in the
liked walking and signature. [2] Nowadays for the person of
automatic finger-print identified systems (AFIS) for stretchable
importance and person of inclination to security of information and stability that liked ending and bifurcation. [5]
in Internet and network. The old tools liked password can not
reply this request for individuals. III. USING VEIN PATTERN IN RECOGNITION
In this moment, we watched creation an electronic commerce,
internet sell and buy of goods, services and based on this The finger-print details must be reminded unchangeable into
transition security is the most importance for companies, banks incision, contusion, black and blue, and the other of damage.
and specifically public people. Therefore the expert persons are Furthermore, many of reason can cause to errors transpire to
searching for the most secure method concerning the security recognize of finger-print liked follow cases:
problem. One way and the succeed way is biometric science. Location of finger of settled, dressing the small part of finger,
Nowadays identification is working based on identifier and sensors of used in sampled and wave sampling.
information personal. But the pattern of grooves and veins in the finger-print is
This method is used to your password personality number individual and the most stability from the others of
(PIN), ID card, keys and the others. specifications. By these descriptions can be profit by those for
Biometric sciences have a new observation based on individual finger-print recognition.
traits
and specials traits that can be protected from changeable, IV. FINGER-PRINT IMAGE AND RECOMMENDATION OF
ALGORITHM
forgetfulness, using with joint. [3] But you know that
finger-print is unchangeable and individual traits. Nowadays
finger-print with these traits can be used into the various A. Finger-print image
connections of identification liked criminology, access control
and internet confirmation.
_____________________________
978-1-4244-4520-2/09/$25.00 ©2009 IEEE


If finger-print is bearing in mind liked the black-and-white
picture, can explain that picture into two-dimension of signals
Set as ( x, y ) as continual local axis.
Also light points of picture of explained into f ( x, y ) .
We can explain the picture of finger-print into f ( m, n)
instead of f ( x, y ) . If we want execute this step, we must
sampled from f ( x, y ) (that defined into continual local axis)
into horizontal and vertical axis like that as f ( m, n) .
Here, we can define finger-print with two dimension of matrix
like M × N . In matrix M and N are make uses pixels of
picture.
Sampling rate of picture is defined the number of pixels in
picture, in the other hand can defined a dimensions of matrix.
The light points of picture is defined a number of light surface
into L. L is defined into variables as 0 and 1 in binary set, and
define into integer variables as 0-255 in gray-scanned photo.
Finger-print image will be shown with binary set of picture
follows:

Figure.2 Finger-print image with part of the binary set of picture


Figure.3 Recommendation of Algorithm

B. Recommendation of Algorithm V. STATEMENT OF FINGER-PRINT ALGORITHM BASED ON


NEURAL NETWORK
With due attention to difference form of veins and grooves in
finger-print, the matrix of image is different. Mat Lab Software is the best proposal for execute this
algorithm, because this Software has a large ability to reading
Also with these explanations we can recognize finger-print
graphic images.
from similar samples. In recommendation method, we must be
Here, we explain of input sample that network will have been
classified finger-print images based on usual method. We can
learning with that. When we send finger-print image to
use Neural Network for recognition of each group. network:
Also we must have used another of neural network for research In step.1 reading Finger-print and will have been making two
into groups.
dimensions of matrix from Finger-print image liked M × N .
For execute this method with neural network must be change
You know that M & N are dimensions of matrix means one
the matrix of finger-print into row-vector. This vector
dimension (M) is row-number and the other one (N) is
(row-vector) is making use of input pattern. column-number. We know that based on IEEE standard
We must learning neural network with different samples of resolution of finger-print images must be 500dpi, 8bpp in our
finger-print based on image matrix. Image matrix is two samples. [6] By this standard of resolution of finger-print image
dimension of matrix that available based on veins pattern and we can show image in size of 512× 512 with 256 points of gray
grooves pattern. on surface. Now M is as 512 and N is as 512 in image vector. A
This method is having a large accuracy to recognition of matrix quantity is as an integer and unsigned value in 8 bit.
finger-print based on veins pattern and grooves pattern. Values of matrix are between 0-255.
Recommendation of algorithm is shown in Figure3 follows: In step.2 must be set beside first pages and second pages are
matrix to each other. Until we can make a new two-dimension
of matrix with m-2n dimension. (m and n are dimension of
finger-print image that discussed above in this part.)
In step.3 must be changing two-dimension matrix into
row-matrix with 1-m-2n dimension.


In step.4 changing precision of vector to multiple precision is as Values of weight are between ( −0.5 ... 0.5 ). Learning rate
double value.
is selected 0.1.
In step.5 we must change dimension of vector is to binary set.
For this act must be dimension of row-vector are divided to 255
Step.2
until we have a vector with 0 and 1 value. For Finger-print images that changing into binary vector, we
Now, if we execute above steps, we can learning neural
must be comes as the sequence of the Step.3 to Step.8.
network. With execute and repeat above steps for all of the (Notation: The changing of image to vector has been defined in
samples in Database and save those into matrix with section of V in my issue).
190-(1-m-2n) dimension can be used those for input pattern of
neural network. (Notation: we used have 190 samples in
Step.3
Database that used those for learning neural network. In this Every learning pattern of neural network for finger-print
issue we discuss about finger-print recognition with the least recognition has an input to input layer. When modifies-weight
error rate.) is computation in input layer, output of this layer as an input
values for headline layer. Then modifies-weight of in headline
A. Selection Neural Network layer is compute.

Step.4
a) Step.1 selection Neural Network for Groups Output of network in headline layer is compute with expression
Recognition of follows:
Here, we must select kind of neural network for Groups 3698
recognition, and will selected Multilayer precpetron with Z − in j = V0 j + ∑ X i .Vi. j (1)
1-m-2n neuron as imported into input layer, 38 neuron as i −1
imported into headline layer, and 19 neuron as imported into Now, Instigation Function of monopoly sigmoid is used to
output layer. Output from network is named of the group. headline layer. At least output of headline layer is computed.
For example: If selected of finger-print image from first Group For computation of output from headline layer must be used
then first output of network is 1, and the others of -1. expression of follows:
Liked this: [1,-1,-1,-1,…,-1] (Notation: Here, Output have a 19
value of row-matrix)
With this act, we can learn Neural Network by imported
Z j = f ( Z − in j ) ( 2)
finger-print image dimension. .
Step.5
Output of network in Output layer is compute with expression
of follows:
b) Step.2 Selection Neural Network for Searching
of Finger-print into Group 30

Now, we must select kind of neural network for searching of Y − in j = W0.1 + ∑ Z j .W j .1 (3)
j −1
finger-print into group where finger-print is classified.
We will be selecting Multilayer Precpetron with 1-m-2n neuron Now, Instigation Function of monopoly sigmoid is used to
as imported into input layer, 20neuron as imported into Output layer. At least output of Output layer is computed.
headline layer, and10 neuron as imported into output layer. For computation of output from Output layer must be used
For example: If requested of Finger-print from first group and expression of follows
we import all of the first value of groups. Network must be
show a row-matrix for first value of 1, and the others of -1. Y1 = f (Y − in j ) ( 4)
Liked that: [1,-1,-1,-1,-1,-1,-1,-1,-1,-1] . (Notation: Here,
Output has a 10 value of row-matrix) Step.6
Note: For learning network, we used Error Propagation. Difference, between output of Neural Network and Output of
B. Learning Neural Network computation is computed. Error rate between System Response
(output of computed of mathematical expressions) and
Computed Response (output of Neural Network), will be
In this section, we defined weight of layers. Input layer with
computed with expression of follows:
dimensions of (1-m-2n)-30 as V in symbol weight.
Headline layer with dimensions of 30 × 1 as W in symbol
weight. δ 1 = (t1 − Y1 ). f ' (Y − in j ) = (t1 − Y1 ).Y1 .(1 − Y1 ) (5)
The Section of learning neural network comes as sequence of
In this expression that show with (5), t1 is an acceptable
the following steps.
response from Neural Network. This will be number of group
Step.1 of the Finger-print. The derivation of Sigmoid Function is
Here, First values of Weight (V and W) are randomizing. computed in expression (6). In the other of expression,
modifies-weight is computed.


f ' ( x) = f ( x).(1 − f ( x)) (6)

ΔW j .1 = α .δ 1 .Z j (7 )

ΔW0.1 = α .δ 1 (8)

Step.7
Here, computing δ for Headline layer.
30
δ − in j = ∑ δ 1 .W j .1 (9)
g −1
Figure.4 Testing Dataset

VII.
δ j = δ − in j . f ' ( Z − in j ) = δ − in j .Z j .(1 − Z j ) (10) CONCLUSION
In this study, we have offered that making use of Neural
Network for searching into image matrix. In this method, we
Step.8 can search and reading image matrix without image processing.
Here, computed the modifies-weight in We can recognize images accuracy.
Headlinelayer. We can recognize Finger-print image easily with accuracy,
ΔVi. j = α .δ j . X i (11) when these image are noisy image, alliterated image, deficient
image, pieced image and the others.

ΔV0. g = α .δ j (12) Acknowledgment


Special Thanks to My Dear Master is Saeed Farzi. Saeed Farzi
is a member of Computer Engineering faculty in Islamic Azad
W j .1 (new) = W j .1 (old ) + ΔW j .1 (13) University Kermanshah-Branch, Iran.
e-mail:saeedfarzi@gmail.com

Vi. j (new) = Vi. j (old ) + ΔVi. j (14)


REFERENCES
[1] http://www.nist.gov/srd/biomet.htm
Step.9 [2] The Biometric Consortium (2002) http:// www.biometric.org
[3] A.K.Jain , et al . , “Biometrics: Promising frontiers for emerging
Can stopped the execute sequence in this time? identification marker” , Computer, vol.33, no.2 , February 2000.
We must be studying condition of pause. [4] Simon-Zorita ,J.Ortega-Garsia , S.Cruz-Lanas and J.Gonzalez-Rodriguez,
“Minutae extraction Scheme for Fingerprint recognition systems’ Image
Processing , 2001 International Conference on , Volume : 3, Oct.2001 .
VI. ACCURACY CONTROL
[5] Marco Gamassi , Vincenzo Piuri , Daniela Sana , Fabio Scotti , “ Robust
We used 256 samples for studying of precision for Accuracy fingerprint detection for access Control “. In worshop Robocare, 2005.
Control. We used 190 samples for learning Neural Network. [6] H.Huang, J.W.Li and W. He , “ A novel Auto-Adaptive Fingerprint
Capture System Based on Image Quality Evaluation Algorithm “.
This notation has been using in section V. The Neural Network IJCSNS , International Journal of Computer Science and Network
has been learning from 190 samples can be recognized all of Security, VOL.6 No.5A, May 2006
them. For Accuracy Control, we changed part of Finger-print
images with software, liked deleting part of Finger-print, but
Neural network is recognized them correctly. We used 95
samples that not to be used for neural network learning. When
we used these for input samples to neural network, neural
network can be recognized 91 samples of them.
Result of this Accuracy control is show, that Error rate of this
method is %96.
Error rate have been 0.000065 for neural network learning.
In Figure.4 can be shown decrease of Error rate of follows:



Вам также может понравиться