International Journal of Computer Science Engineering and Information Technology Research (IJCSEITR) ISSN(P): 2249-6831; ISSN(E): 2249-7943 Vol.

4, Issue 1, Feb 2014, 103-110 TJPRC Pvt. Ltd.

FINGERPRINT RECOGNITION SYSTEM USING SUPPORT VECTOR MACHINE AND NEURAL NETWORK
SOZAN ABDULLA MAHMOOD Department of Computer Science, Sulaimany University, Sulaymaniyah, Iraq

ABSTRACT
Verification is reliable personal identification method which play a very important role in different applications. The main objective in this paper, the geometrical shapes are used to extract features based on minutiae point. The features have been used as a set of descriptors for the fingerprint data and it is too complex to reconstruct the original fingerprint image using the extracted data. Back propagation Neural Networks and Support Vector Machine were used for Classification and recognition in the purposed system. It is found that the process of preprocessing steps necessary for accurate minutiae extraction including median filter, binarization , and thinning. Alsothe method of constructing geometric shapes has great effectson producing good results in the recognition rate. The recognition rate of the Neural Network is more accurate than Support Vector Machine.

KEYWORDS: Fingerprint Recognition, Support Vector Machine, Back propagation Neural Network, Minutiae
Extraction

1. INTRODUCTION
Fingerprint is the most widely used biometric system and have a very important role in forensic and civilian application, so it will continue to be used with many governments legacy systems. The most widely and well-known used method for fingerprint is minutiae extraction method.Minutiae of fingerprint include ridge bifurcations, ridge ending, short ridge and enclosure in D. Maltoni, et al. (2003). In Terje (2012) two different classification regimes are used to train the systems to do correct classification. The classification rate has been estimated to about 87.0 % and 88.8% of unseen fingerprints for SVM and MLP classification respectively. In Mohamed Mostafa (2005) the clustering algorithm and new supervised recurrent neural-network is used to detect similar features groups from multiple template images that generated from the same finger. In Antowiakand Chalasinska(2003) the optical wavelet transform used as a preprocessor for an artificial neural network. The achieved capabilities include limited shift-, rotation, scale- and intensity-invariance, also edge- enhancement filter performed to improves the ability of the system. Mori and maltoni (1998) presented an approach to minutiae filtering based on a neural network. The minutiae neighborhoods extracted and normalized with respect to rotation and scale. They employed the neural classifier to perform the neighborhoods classification, which topology has been designed to exploit the minutiae duality. In this paper, extracting minutiae (bifurcation and ridge ending) from an input fingerprint image and produce a feature set by using geometry shape. Using the extracting feature to feed to neural network. Section II in this paper, concerns with the main fundamental concept to understand the idea of recognition system using fingerprint and fundamental about extracting feature from minutiae. Also, in this paper the properties of neural network and Support vector machine were discussed. Section 3 introduce the flow of the enrollment and identification phase of proposed identification algorithm. Then, the preprocessing and minutiae extraction are represent, recognition approaches was described and how

104

Sozan Abdulla Mahmood

the system databases store the template record of all individuals that have access to the system. In, section 6. the experimental results were discussed in detail, the compression of two classification methods were explained.Finally, Section 7.summarizes the conclusions.

2. MATCHING FINGERPRINT MINUTIAE

Fingerprint matching techniques can be placed into two categories: minutiae-based and correlation based. Minutiae-based techniques first find minutiae points and then map their relative placement on the finger. While correlation-based method is able to overcome some of the difficulties of the minutiae-based approach. However, it has some of its own shortcomings. Correlation-based techniques require the precise location of a registration point and are affected by image translation and rotationinin Salil and Anil (2000). A fingerprint image differs from scan to scan by some combination of two dimensional translation, rotation, and scaling; and a three dimensional rolling, which occurs when the user does not place his finger at the same elevation fro m scan to scan in Amendt, and Friend(2001). Many fingerprint classification methods have been proposed. In general, these methods can be catego-rised into five approaches in Terje (2012) . rule based syntactic based structure based statistical based neural network based Artificial Neural Networks have been applied to many problems, and have demonstrated their superiority over classical methods when dealing with noisy or incomplete data. One such application is the authentication. This extracted information (stored in a hidden layer) preserves the full information obtained from the external environment. In this paper, the first classification was backpropagation neural network, which was adopted since it has been successfully applied to many pattern classification problems including authentication. The structure of the backpropagation neural network consists of three layers: first layer input neurons which are fully connected to the hidden layer. The last layer is the output layer consisting of 5 neurons whose output uses to binary encoding ID of persons fingerprint. All three layers are fully feed forwarded trained. The activation function, where the unipolar sigmoid function is used. Also, SVMs have been successfully applied to various classification problems, which perform pattern recognition for two-class problems by determining the separating hyperplane with maximum distance to the closest points of the training set in Chaohongand Venu (2004). The SVM is used to transform the original data set to ahigher-dimensional or feature space by using different kernel functions can solve non linear problems. Figure 1 shows the support vector machine with kernel function in Ashish (2012).

Fingerprint Recognition System Using Support Vector Machine and Neural Network

105

Figure 1: Support Vector Machine with Kernel Function

3. PROPOSED APPROCH
The proposed system as shown in Figure 2 consist of three main steps: preprocessing, feature extraction and (testing and training). The first step is removing noise from the image may be corrupted during the fingerprint image capture, using binarization technique to convert gray-scale to black and white image, after that confirm mean filter and thinning on an image. The second step extracting minutiae, constructing triangle shape from the minutiae point and the feature stored in a database as a template. In the last step the extracted feature, fed forward back propagation neural network and support vector machine were trained and caparison between two different classification methods were done.

Figure 2: Illustrates the Layout of the Proposed Authentication System Based on Fingerprint

4. PREPROCESSING
The initial step in the proposed authentication system is computing the binary image from the input gray scale fingerprint image, by applying threshold value. As the image may have various grey-level value due to non-uniformity of the ink intensity, non-uniform contact with the sensors by users or changes in illumination and contrast during image acquisition process. Adaptive threshold can be used to binaries fingerprint images, binarization depends on the comparison result of grey-level value of each pixel with local meaninGovindaraju (2006):

1 if f ( x, y )
IB(x,y)

1 / W f ( x i, y j )
i 1 j 1

= . 0 Otherwise

(1)

106

Sozan Abdulla Mahmood

Where f(x,y) is a gray-levelat (x,y), and W is the widow size, the black pixels are denoted by zero while white pixels are represented by one. After binarization, another major preprocessing technique for image is implemented which is the thinning method is often called skeletonization to get an image of single pixel width with no discontinuities, each ridge is thinned to its centre pixel, and singular pixels are eliminated.As it shown in Figure 3.

(a)

(b)

Figure 3: (a) Original Image in Gray, (b) Thinning Black-White Image

5. MINUTIAE EXTRACTION
Most fingerprint minutia extraction methods are thinning- based where the skeletonization process converts each ridge to one pixel wide. Minutia points are detected by locating the end points and bifurcation points on the thinned ridge skeleton based on the number of neighboring pixels inJain et al. (1997). The concept of Crossing Number (CN) is widely used for extracting the minutiae in Govindaraju (2006). The crossing number for a pixel P is: P4 P5 P6 P3 P P7 P2 P1 P8 (2)

CN 1 / 2 Pi Pi 1 i 1
where Pi is the binary pixel value in the neighborhood of P with Pi = (0 or1) and P9=P1.

The skeleton image of fingerprint is scanned and all the minutiae are detected using the following properties of CNas shown in Table 1: Table 1: Crossing Numbers Property
CN 0 1 2 3 4 Property Isolated point Ending point Connective point Bifurcation point Crossing point

Then depending on the value of CN the minutiae point which have ending orbifurcation property are taken into consideration. To calculate direction of both pointsthe following code is used: Suppose P is the checked point and P1 to P8 are neighborhood pixels

Fingerprint Recognition System Using Support Vector Machine and Neural Network

107

if CN =1 then if P1 = 1 then D=W elseif P3=1 then D=S elseif P7=1 then D=N elseif P5=1 then D=E elseif P4 =1 then D=SE elseif P2=1 then D=SW elseif P6=1 then D=NE elseif P8=1 then D=NW endif if CN =3 then if P1 &P3& P7 = 1 then D=W elseif P1 &P3& P5=1 then D=S elseif P1 &P7& P5=1 then D=N elseif P3 &P5& P7=1 then D=E elseif P4 &P3& P5=1 then D=SE elseif P3 &P2& P1=1 then D=SW elseif P3 &P5& P6=1 then D=NE elseif P4 &P8& P5=1 then D=NW endif In the proposed method, the minutiae points locations, and their considereddirection from the 8 directions (N,S,W,E,NE,NW,SE,SW) are recorded then they used to construct triangle shape depending of the number of recorded minutiae point and their direction. The next step is calculating the area of each triangle, after getting the triangle area values, they are sorted in descending order, the first 10 areas will be stored in database which representing of fingerprint image. So in this case concentrates on reducing the computation time for feature extraction and matching stage. In most embedded fingerprint systems supporting instant identification or verification are increasingly used, and the computation time for these processes is thus an important research field. To decrease this time is to take part of feature will be trained in neural network or SVM, such that only a part of the fingerprints feature s are taken for matching. In this paper, the skeleton image of the fingerprint has been scanned, and all types of points in the image depending on the CN values from the Table(1) has been found. In this paper only Ending point and Bifurcation point are considered. The location (x,y), and the direction of each point with CN value equal to (1 or 3) were recorded in the matrix. For each point P(i,j), the proposed algorithm will select two points from the recorded points depending on the

108

Sozan Abdulla Mahmood

direction of the point P(i,j), for example if considered direction is N for point P so the searching process is done for two points located at (x,y) where x<i also for NE and NW taken the same points, while the considered direction is S, SE, and SW the searching process is done for point located where x>i then the distance calculated between P and the another two points in order to be suitable for creating the triangle , this process will be repeated for all the two points that have the above condition. The following two equations are used to find the distance between the points and area of triangle:

D ( y 2 y1) 2 ( x 2 x1) 2

(3)

TArea

1 AB sin 2

(4)

Where A and B is distance between P(i,j) and the two points , a is angle between A and B

6. EXPERIMENTAL RESULTS
To evaluate the performance of our algorithms, we randomly select 380 sample, for each subject(3) fingerprint images from our fingerprint database which consist of 560 fingerprint image for 40 persons. In the experiments, the scanned fingerprint with size (170 x170). In the proposed system the minimum 10 areas of triangles for each fingerprint image are taken and used in training stage, with different number of hidden nodes, the weight are adjusted,and the output is the binary ID number for a person represented by 5 output nodes. The effectiveness of proposed algorithm is shown by the improvement in fingerprint verification accuracy, comparing the two method neural network and SVM. The recognition rates for training neural network are all approaching 100%. Whilerecognition rate in percentage on testing datais about %92, and in Support Vector Machine algorithm is % 85, the effect of numbers of triangles area value for triangles and the number of hidden nodes in recognition rate are shown in Table 2 and Table 3. Table 2: The Effect on Hidden Node on Recognition Rate No. of Input Node 6 8 10 Recognition Rate % Neural Network System Hidden 5 Hidden Hidden Nodes 7 Node 10 Nodes 79 83 84 80 87 87 84 92 90

Table 3: The Recognition Rate Using SVM Cross Validation Rate 75% 80% 86% Recognition Rate % SVM System 70% 77% 89%

7. CONCLUTIONS
Fingerprint minutiae and a neural network with back propagation training algorithm are appropriate to use for an authentication system. The proposed system which constructing the triangle shapes from the extracted feature based on minutiae point and considered direction, have a good effective and efficient in recognition system. Improving the accuracy

Fingerprint Recognition System Using Support Vector Machine and Neural Network

109

and the ability of the proposed system to get more minutiae points can be done by performing and testing various enhancement filter to the fingerprint image. The experimental results have shown that the best average recognition rate for Neural Network is 92% with learning rate is 0.01 , 7 hidden nodes, and 10 input node shows a great powerful recognition performance, while the recognition rate for Support Vector Machine is 89% with cross validation is 8 and value is 0.00564.

REFERENCES
1. Ashish B. I, Dr. D. S. C Speech Emotion Recognition Using Hidden Markov Model and Support Vector Machine, IJAERS/Vol. I/ Issue III/April-June, 2012/316-318 2. Chaohong Wu, Zhixin Shi, and VenuGovindaraju, Fingerprint Image Enhancement method Using Directional Median Filter, Elsevier Science, 2004, pp. 8-9. 3. 4. D. Maltoni, D. Maio, A.K. Jain, S. Prabhakar, Handbook of Fingerprint Recognition, Springer Verlag, 2003. J. Bigun and F. Smeraldi (Eds.): Fingerprint Classification with Combinations ofSupport Vector Machines AVBPA 2001, LNCS 2091, pp. 253258, 2001. 5. 6. K. Amendt, D. Friend , Fingerprint Identification, MIT Course 6.111: Digital Electronics Lab, 2006. K. Jain, L. Hong and R. Bolle, "On-Line Fingerprint Verification'', IEEE Trans. Pattern Anal. Machine Intell., Vol. 19, . No. 4, 1997, pp. 302-314. 7. Mohamed MostafaAbd Allah Artificial Neural Networks Based Fingerprint Authentication with Clusters Algorithm, Informatica Vol. 29, 2005, pp 303307. 8. M. Antowiak and K. Chalasinska-Macukow, Fingerprint Identification by Using Artificial neural network and Optical Wavelet Preprocessing, OPTO-ELECTRONICS REVIEW 11(4), 2003, pp 327-337. 9. Maio, D., Maltoni, D., Neural network based minutiae filtering in fingerprints IEEE Xplore, 1998.

10. SalilPrabhakar, Anil K. Jain, Learning Fingerprint Minutiae Location and Type International Conference on Pattern Recognition(ICPR), 2000. 11. TerjeKristensen Two Different Regimes of Fingerprint Identification a Comparison, American Journal of Computational and Applied Mathematics 2012, 2(2):pp 1-9 12. Z. Shi, V. Govindaraju A chaincode based scheme for fingerprint feature extraction, Patte rn Recognition Letters 27, 2006, pp 462468.