Offline Signature Verification Using Pixel Matching ...

17 downloads 675 Views 403KB Size Report
Offline Signature verification is an authentication method that uses the dynamics of a person's handwritten signature measure and analyses the physical activity of signing. The core of ... Available online at www.sciencedirect.com. © 2013 The ...
Available online at www.sciencedirect.com

ScienceDirect Procedia Technology 10 (2013) 970 – 977

International Conference on Computational Intelligence: Modeling Techniques and Applications (CIMTA) 2013

Offline Signature Verification Using Pixel Matching Technique Indrajit Bhattacharyaa*, Prabir Ghoshb, Swarup Biswasb a

Department of Computer Applications, Kalyani Govt. Enginooring College, Kalyani, Nadia-741235, West Bengal, India b Deaprtment of Computer Application, Santipur College, Nadia, West Bengal, India

Abstract Biometrics (or biometric authentication) refers to the process of identification of humans by their characteristics or traits. Biometrics is used in computer science as a form of identification and access control which is one of the most secure methods to keep humans privacy. Biometric can be classified into two categories: behavioural (signature verification, keystroke dynamics, etc.) and physiological (iris characteristics, fingerprint, etc.). Handwritten signature is amongst the first few biometrics to be used even before the advent of computers. Offline Signature verification is an authentication method that uses the dynamics of a person’s handwritten signature measure and analyses the physical activity of signing. The core of a signature biometric system is behavioural, and in this paper we have proposed an off-line signature verification and recognition system using pixel matching technique. PMT (Pixel Matching Technique) is used to verify the signature of the user with the sample signature which is stored in the database. The performance of the proposed method has been compared with the existing ANN (Artificial Neural Network’s) back-propagation method and SVM (Support Vector Machine) technique. © 2013 2013 The The Authors. Authors.Published PublishedbybyElsevier ElsevierLtd. Ltd. Open access under CC BY-NC-ND license. Selection thethe University of of Kalyani, Department of Computer Science & Engineering. Selection and and peer-review peer-reviewunder underresponsibility responsibilityofof University Kalyani, Department of Computer Science & Engineering Keywords: Signature, PMT, Authentication, Biometric Authentication

1. Introduction A signature is composed of special characters and flourishes and uses to authentication one human beings from another. In this authentication process a captured signature is stored in a computer in the form of image file. The problem is to compare the user signature with a sample database signature.

* Indrajit Bhattacharya Tel.: +91- 9830051359; fax: +033-2548-8362. E-mail address: [email protected]

2212-0173 © 2013 The Authors. Published by Elsevier Ltd. Open access under CC BY-NC-ND license. Selection and peer-review under responsibility of the University of Kalyani, Department of Computer Science & Engineering doi:10.1016/j.protcy.2013.12.445

Indrajit Bhattacharya et al. / Procedia Technology 10 (2013) 970 – 977

Signature Authentication is a very important process. It finds its application in many sectors like Banking, Property Dealing, and other areas. If we can develop any software or device like SAM (Signature Authentication Machine) then it can be applicable in many exiting system. Signature verification contains two areas: off-line signature verification, where signature samples are scanned into image representation and on-line signature verification, where signature samples are collected from a digitizing tablet which is capable of pen movements during the writing. In our work we have compared the offline signature of a person with the sample signature stored in database. 2. Related works In 2009, Ghandali and Moghaddam have proposed an off-line Persians signature identification and verification based on image registration, DWT (Discrete Wavelet Transform) and fusion. The authors used DWT for Features extraction and Euclidean distance for comparing features. It is language dependent methods [1, 9]. In 2008, Larkins and Mayo have introduced a person dependent off-line signature verification method that is based on Adaptive Feature Threshold (AFT). AFT enhances the method of converting a simple feature of signature to binary feature vector to improve its representative similarity with training signatures. They have used combination of spatial pyramid and equimass sampling grids to improve representation of a signature based on gradient direction. In classification phase, the authors used DWT and graph matching methods [2]. In 2007, kovari et.al presented an approach for off-line signature verification, which was able to preserve and take usage of semantic information. The authors used position and direction endpoints in features extraction phase [5]. Alan McCabe et al. proposed a method for verifying handwritten signatures by using NN architecture. Various static (e.g., height, slant, etc.) and dynamic (e.g., velocity, pen tip pressure, etc.) signature features are extracted and used to train the NN Several Network topologies are tested and their accuracy is compared [14]. A Baysian model for off-line signature verification involving the representation of a signature through its curvature is developed by McKeague [12]. Ferrer et al. calculates geometric features of a signature in fixed-point arithmetic for offline verification. The proposed features are then checked with different classifiers, such as Hidden Markov Models, Support Vector Machines etc. [13]. A novel approach to off-line signature verification is proposed by Wei Tian et al. Both static and pseudo dynamic features are extracted as original signal, which can enhance the difference between a genuine signature and its forgery [11]. Fang et. Al. proposed two methods for the detection of skilled forgeries using template matching. One method is based on optimal matching of the one dimensional projection profiles of the signature patterns and the other is based on elastic matching of the strokes in the two dimensional signature patterns [10]. Justino [3] et al. in his work presented a robust system for offline signature verification using simple features, different cell resolutions and multiple codebooks in an HMM framework. The simple and random forgery error rates have shown to be low and close of each other. There are many other existing algorithms proposed by the researchers to solve signature verification problem like CSMOSV, MDF and RBF etc. 3. Proposed method The proposed system is divided into two major phases. I. Pre-processing, II. Verification: I. Pre-Processing a. Capturing signature from the page b. Removing noise and colour c. Adjust the property of scanned signature i. Find the exact signature in the signature box ii. Find the angle of the signature and rotate it with that angle iii. Resizing the signature with the size of sample database signature

971

972

Indrajit Bhattacharya et al. / Procedia Technology 10 (2013) 970 – 977

II.

Verification

The block diagram of the proposed system is given below in Figure 1.

Scanned Signature Image

Pre-processing

Capturing Signature

Noise & Colour Removal

Angular Rotation

Adjust Property

Exact Position detection

Resizing

Verification

Fig. 1. Block diagram of the proposed scheme

4. Description of the Proposed Scheme 4.1. Capturing signature from a page Normally a scanner scans a page as an image. Here the problem is to identify the signature from a scanned image. It is difficult because there can be many other text and patterns in that scanned image. Hence it is required to set a signature area, which will help to identify the exact boundary of the signature in that scanned image. If the signature area had been identified (where users endorse their signature) bounded with a border color (like RGB-239, 228, 176) then it would help to identify the exact position of that signature in that scanned image. The image containing the user signature has been scanned using the following algorithm:

Indrajit Bhattacharya et al. / Procedia Technology 10 (2013) 970 – 977

4.1.1. Algorithm to calculate the rectangular signature area: Step.1 Set st = 0, sl = 0, sh = 0, sw = 0 (sl is signature left position an st is top sh is Height and sw is width) Step.2 Set x = 0 Step.3 Set y = 0 Step.4 Scan the color from the scanned image and store it in pixelcolor Step.5 Check if pixelcolor = RGB-239,228,176 then goto Step 6 else goto Step 8 Step.6 Check if sw = 0 then Set sl = x AND if sh = 0 then Set st = y Step.7 Set sw = x; sh = y; y = y + 1 Step.8 Check if y < image.height - 1 then Repeat from Step 4 to Step 7 Step.9 Set x = x +1 Step.10 Check if x < image.width - 1 then Repeat from Step 3 to Step 9 Step.11 End 4.2. Removing noise and normalize the color Color and noise removal is very important because a signature may compose of many colors and may be affected by noises after scanning. Hence we have to eliminate all the noise present across the signature regional area to get the exact signature. After noise elimination the image had been converted to a black & white image. This can stored in the database as a sample signature or can be used to compare with a sample database signature. The advantage is that it decreases the size of the image and it is required to compare only two colors. In order to solve this problem the following methods have been proposed. 4.2.1. Color normalization method To make the rectangular area black and white we need to scan all the rectangular area using the below algorithm and find the color of each pixel. If the color is (RGB (239,228,176)) then converts white and if the color is white then there is no conversation. Otherwise change the pixel color to black. This is so easy and fast technique to make a black and white picture. The technique of this is given below as algorithm. 4.2.1.1. Algorithm to make the image black and white: Step.1 Scan the color from the scanned image and store it in pixelcolor Step.2 If pixelcolor = color (rgb (239,228,176)) then Set pixelcolor = color (white) Step.3 If pixelcolor = color (white) then goto Step 4 else Set pixelcolor = color (black) Step.4 Repeat from Step 1 to Step 3 while image not scanned completely Step.5 End 4.2.2. Proposed noise resolution method In the next step of the proposed work the noise has to be removed. In this phase the noise associated with the image which has been created by the scanner while the image has been scanned need to be removed. After scanning that signature and normalizing its color some small single pixels of black color has been found, which is not the part of that signature. Following method has been introduced to remove the noise in the image. 4.2.2.1. Algorithm to remove noise: Step.1 Scan the color from the scanned image and store it in pixelcolor Step.2 If pixelcolor = color (black) then goto Step 3 else goto step 4 Step.3 If pixelcolor is not same wit adjoin pixelcolor then set pixelcolor = color (white) and goto Step 1 else goto step 4 Step.4 Repeat step 1 to step 3 while image not scanned completely Step.5 End

973

974

Indrajit Bhattacharya et al. / Procedia Technology 10 (2013) 970 – 977

4.3. Adjust its property After the previous step a binary image (only black and white) is obtained. After that it is required to locate the exact position of the signature in the image to perform the signature verification, because signature can be anywhere inside this rectangular area. And it can sign in different angles and sizes. Now the first problem is to find the exact position of the signature from the rectangular boundary area and the second problem is to find the angle and size of the signature. After that the necessary correction need to be done in the context mentioned above. 4.3.1. Finding the exact position of the signature in the signature box To solve the first problem a solution has been proposed. The proposed solution is based on the identification of edges of the signature in the signature box. It scan the rectangular area using the above algorithm 4.1.1 from its four (i.e. top, bottom, left, right) sides. And after that the actual signature area has been extracted.

Fig. 2. Finding the exact position of the signature in signature box

4.3.2. Angular problem solutions Another important task is that the angular detection of a user signature can change from time to time. A signature can be written in different angles. To compare the signature signed in different angles that is stored in database. The main goal is to change the property of scanned signature such that it can be compared with the sample database signature. Sometime same signature can be written in an angle (Fig.3.a. Original Image, Fig.3.b. Angular Image), and to solve this angular problem some mathematical formula is used that has been described below.

Fig. 3. (a) Original signature; (b) Sample signature

We have used co-ordinate geometry to find the angle and to rotate the image accordingly. An angular signature which can be described by Fig.4 can be considered. In this picture two points ‘a’ and ‘b’ are the two end points of the signature.

Indrajit Bhattacharya et al. / Procedia Technology 10 (2013) 970 – 977

975

Fig. 4. An angular signature

Now we get x1, y1 co-ordinates from point ‘a’ and x2, y2 co-ordinates from point ‘b’. If we draw a straight line to ‘w’ and ‘h’ from point ‘a’ and ‘b’ respectively then we can see that the two lines from ‘a’ and ‘b’ cuts each other at point ‘c’. And now from the co-ordinate geometry, the angle is ℎ = ℎ ℎ ∴ = tan ℎ

ℎ = − ; ℎ = − In order to measure the angle properly the angle θ has been scaled 100 times. Then the image has been rotated by the angle θ to make it horizontal to the x axis. 4.3.3. Resizing image with database image After rotation the signature as the signature size gets enlarged hence resizing of the signature is required before the two signatures get compared. 4.4. Compare the signature with database signature To compare the signature with the sample database signature a simple algorithm has been introduced based on pixel matching concept. It is easy to implement and computationally less complex. Algorithm 4.1.1 is proposed to scan every pixel color of the scanned image and the sample database image. Scan the two images from left to right check for the pixel color. If the pixel color is black then it shows that it is a part of signature and compares the corresponding pixel with the other image. If the pixel color is same for two images than a counter (m) increased. And if the match is not found than another counter (n) is increased. By scanning every black pixel also a counter (p) increased. The percentage of matching [8] is calculated by the following formula: =

+

×

976

Indrajit Bhattacharya et al. / Procedia Technology 10 (2013) 970 – 977

4.4.1. Algorithm to compare the signature with sample database signature Step.1 Set m = 0, n =0, p = 0 Step.2 Scan the color from the database image and store it in pixcelcolor Step.3 Scan the color from the scanned image and store it in color Step.4 If pixelcolor = color (black) then step 5 else step 7 Step.5 If pixelcolor = color then Set m = m + 1 else Set n = n + 1 Step.6 Set p = p + 1 Step.7 Repeat Step 2 while image not scanned totally Step.8 Set p = m / (n + p) *100 Step.9 Show the value of p 5. Comparison with another method In this system for each person 8 original and 8 forgery signatures are tested. The possible cases in verification are true acceptation (TA), false rejection (FR), true rejection (TR), false acceptation (FA). The verification results of this system and SVM and ANN’s Back propagation [4, 6, 7], method are given in Table 1. Table 1. Comparison of verification results System

TAR

FAR

FAR

FRR

PMT

0.94 0.98 0.78

0.06 0.02 0.22

0.88 0.89 0.84

0.12 0.11 0.16

ANN SVM

From the above result it can be concluded that the performance of our proposed method is comparable with the existing schemes. But the advantage of the new technique is that it is very simple and easy to implement. It can be applicable to most of the Applications where offline Signature authentication is required. 6. Conclusion Signature authentication machine is implemented to provide a simple, safe, fast biometric behavioral security system. By using some equations from co-ordinate geometry makes this method faster than other methods. The color matching technique makes it more secure. The Interface of this application is very simple which makes it user friendly and easy. The implemented system has the following limitations:  Our system only able to identify the static changes in a signature, it cannot identify any dynamic changes in a signature.  This signature authentication system works offline. No online device is connected with this system. And the methods of signature authentication tested only offline.

References [1]

S. Ghandali and M. Ebrahimi Moghaddam, Off-Line Persian Signature Identification and Verification based on Image Registration and Fusion, In: Journal of Multimedia, volume 4, 2009, pages: 137-144.

[2]

Larkins, R. Mayo, M., “Adaptive Feature Thresholding for Off-Line Signature Verification”, In: Image and vision computing New Zealand, 2008, pages: 1-6.

[3]

E. J. R. Justino , F Bortolozzi and R. sabourin , Off-line signature verification Using HMM for random, simple and skilled forgeries, ICDAR 2001, International Conference on document Analysis and Recognition, vol. 1 pp. 105-110. 2001.

[4]

B. zhang, M. Fu and H. Yan, Handwritten signature Verification based on Neural ‘”GAS” based Vector Quantization, IEEE

Indrajit Bhattacharya et al. / Procedia Technology 10 (2013) 970 – 977

International Conference on neural networks, pp. 1862-1864,May, 1998. [5]

Kovari, B. Kertesz, Z. and Major, a., Off-Line Signature Verification Based on Feature Matching: In: Intelligent Engineering Systems, 2007, pages 93-97.

[6]

A. chalechale and A. Martins , Line seament distribution of sketches for Persian signature Recognition, IEEE Proc. TENCON, vol. 1 pp.11-15 Oct, 2003.

[7]

C. Cortes and V. Vapnik Support-vector networks, Machine Learning, vol 20, pp, 273-297, nov, 1995.

[8]

V.N. Vapnik The nature of Statistical Learning Theory, springer, 1995.

[9]

Meenakshi S Arya, Vandana S Inamdar, A Preliminary Study on Various Off-Line Hand Written Signature Verification Approaches.

[10] B.Fang, C.H Leung, Y.Y. Tang, K.W. Tse, P.C.K. Kwok, and Y.K.Wong, Off-line signature verification by the tracking of feature and stroke positions, Pattern Recognition, vol. 36, pp. 91-101, 2003. [11] Wei Tian, YizhengQiao and Zhiqiang Ma, A New Scheme for Off-line Signature Verification Using DWT and Fuzzy Net”, 8th ACIS International Conference on Software Engineering, Artificial Intelligence, Networking, and parallel/Distributed Computing. [12] Ian W. McKeague, A Statistical model for signature verification, Journal of the American Statistical Association, 2005, vol. 100, pages231-241. [13] Miguel A.Ferrer, Jesu’s B. Alonso, and Carlos M. Travieso, Offline Geometric Parameters for Automatic Signature Verification Using Fixed-Point Arithmetic IEEE transactions On Pattern Analysis And Machine Intelligence, vol. 27, No. 6, June 2005 [14] Alan McCabe, Jarrod Trevathan and Wayne Read, Neural Network-based Handwritten Signature Verification, Journal of computers, vol. 3, no. 8, August 2008.

977