You are on page 1of 5

IPASJ International Journal of Information Technology (IIJIT)

Web Site: http://www.ipasj.org/IIJIT/IIJIT.htm


A Publisher for Research Motivation ........ Email:editoriijit@ipasj.org
Volume 7, Issue 5, May 2019 ISSN 2321-5976

Offline Signature Verification using LBP


Technique
G.G.Rajput1, Panditkumar Patil2
1
Department of Computer Science, Karnataka State Women’s University, Vijayapur, Karnataka, India
2
PDepartment of Computer Science, Gulbarga University, Kalaburgi, Karnataka, India

Abstract
Signatures are widely used approach for verifying the identity of a person in many applications including credit card validation,
banking security systems etc. Offline signature verification is done on the scanned image of handwritten signature. In this
paper, we propose offline signature verification system using features extracted based on Local Binary Pattern (LBP) approach.
The LBP is computed on pixel by pixel basis by defining local structure information with its adjacent pixels. K-NN and SVM
classifiers are used for verification of the signature. Experiments are performed on the signature images of MCYT dataset and
local database.
Keywords: biometric, LBP, signature verification, KNN, SVM.

1. INTRODUCTION
Now a day’s biometric based identification plays very important role in applications like person authentication for
performing business transactions, security issues, etc [1]. Hand written signature is most widely accepted biological
feature. Signature verification can be performed on-line or off-line. In Online verification, a signer signs on electronic
devices such as Tablet PC, touch screen with an electronic pen, and features like pressure exerted, stroke length,
writing speed are captured for the authenticating the signature[2]. In offline signature verification, the handwritten
signature appearing in the document image is extracted into digital form (Figure 1) and then signature verification is
accomplished by comparing the signature with the template signature stored in the database [3].

Figure1. Sample signature images

Signature verification is a crucial task, since a person’s signature may be forged with intent to receive the person’s
identity for financial gain (eg. forging a signature on bank cheques, stock certificates), legal claim (eg. deeds, patents),
business transactions, etc. In case of forgery, a person tries to copy the signature of another person. The forgery
signature can be classified as random forgery, simple forgery, and skilled forgery [3], [4]. Authentication of the
signature in case of skilled forgery is a challenging task compared to other category of forgeries because the signer
(false person) knows the signature as well as the pattern very well and has experience in forgery. The signer signs
exactly the same as the victim and it is very difficult to compare between original and forged signature [4].

Methods are found in the literature with different approaches towards feature extraction and use of different
classification models for verifying the genuinely of the signature. A brief review of literature for offline handwritten
signature identification is presented below.
A review of recent advances in signature verification is presented by Luiz G. Hafemann et. al. [5]. Shashi Kumar et. al
[6] presented a grid based approach using global features for off-line signature verification. The skeletonized image is
divided into 120 rectangular segments and signature pixels are counted for each block. Also global features namely,
aspect ratio, signature height, image area, and width and height are computed. Multilayer feed forward artificial neural
network is used to train the features and authenticate the test signature image. M.S. Shirdhonkar et.al.[7] proposed to
use the rotated complex wavelet filters (RCWF) and dual tree complex wavelet transform(DTCWT) together to derive
signature feature extraction, which captures information in twelve different directions. In identification phase,
Canberra distance measure is used.
A scale and rotation invariant technique for signature identification is proposed by Supinder Singh and Amandeep

Volume 7, Issue 5, May 2019 Page 17


IPASJ International Journal of Information Technology (IIJIT)
Web Site: http://www.ipasj.org/IIJIT/IIJIT.htm
A Publisher for Research Motivation ........ Email:editoriijit@ipasj.org
Volume 7, Issue 5, May 2019 ISSN 2321-5976

Kaur[8], which involves extraction of invariant rotation invariant texture features (sub uniform local binary patterns)
from the 12 blocks within an image. DCT coefficients are used for verification. Identification and verification of the test
signature are performed with the combination of chi square distance and support vector machines. P. Vikram et. al. [9]
proposed the Support Local Binary Pattern (SLBP) features for offline signature verification. The feature vectors were
evaluated using k- nearest neighbor and Linear Discriminant Classifiers.

From the literature review, it is evident that numerous methods are presented performing experiments on either local
dataset or standard dataset like MCYT. Variations of LBP are implemented in signature verification process by
different authors. Signature verification remains a challenging task and that researchers are continuing to explore better
feature representations to improve the classification. In this paper, we present efficient method for signature
verification using LBP in its basic form. SVM and KNN classifiers are chosen for signature verification. The efficacy of
the method is demonstrated in terms of experiments performed on local dataset of signatures as well as signatures from
MCYT database. The process model is shown in figure 2.

Figure 2. Verification process

2. DATA COLLECTION AND PREPROCESSING

MCYT bimodal database consists of large collection of fingerprint and signature images [10]. MCYT signature
database contains signatures of 75 users with 30 signatures per person which includes 15 genuine and 15 skilled
forgeries. Random selection of signature images has been considered in this paper for training and testing purpose.
The details of creating local database of offline signatures are presented below.

The purpose of creating local dataset of signatures is to consider the signature images drawn in local languages.
Further, it is observed that the signature drawn by a person in English is also influenced by the local language he/she is
using for communication. Handwritten signatures are collected from the persons with different age groups and
professions. A 30 subjects of 16 each total 480 handwritten signatures are collected. The collected handwritten
signatures are scanned using HP flatbed scanner. Details of the signatures collected are presented in table 1.

Table 1. Statistics of signatures collected


Subject distribution Female: 30% Male: 70%
Subject age distribution 18-60
subject category employee, student, business person

The scanned images of gray scale are then converted to binary images by applying Otsu’sthresholding method. Median
filter is applied to remove salt and pepper noise.

3. FEATURE EXTRACTION AND CLASSIFICATION


LBP operator [11], [12] is used for feature extraction. LBP is a gray-scale invariant texture measure, derived from a
general definition of texture in a local neighborhood. LBP operator produces a binary code by thresholding a 3x3
neighborhood by the gray value of its center. The histogram of these labels can then be used as a texture descriptor.
LBP operator is robust to monotonic gray-scale changes caused and it has the advantage of simple implementation
when compared to other techniques like SIFT [13], [14].

The BASIC procedure for computing the binary code is as follows. In its basic form of implementation, LBP operator
takes 3 x 3 neighborhood of a pixel and generates a binary 1 if the neighbor of the center pixel has the larger value than
the center pixel. The operator generates a binary 0 if the neighbor is less than the center. For 8-neigborhood, an 8 digit
binary number is generated which is represented as unsigned integer, making it a compact description. An example of
binary code generation is shown in figure 3.

Volume 7, Issue 5, May 2019 Page 18


IPASJ International Journal of Information Technology (IIJIT)
Web Site: http://www.ipasj.org/IIJIT/IIJIT.htm
A Publisher for Research Motivation ........ Email:editoriijit@ipasj.org
Volume 7, Issue 5, May 2019 ISSN 2321-5976

Figure 3. Calculating the LBP code

Feature extraction algorithm is presented below.


Input: Pre-processed signature images in binary
Output: LBP feature vector
1. Divide the image into 3x3 cells and for each cell compute the compact representation of LBP operator generated
binary code.
2. Compute the histogram of each cell with combination of pixel smaller and greater than the centre pixel value.
3. Normalize the histogram to neighborhood pattern of (8, R), where R=1.
4. Feature vector of entire window is obtained by concatenating histogram of each cell having 59 uniform pattern
(out of 256).

Sample LBP image for signature image is shown in the figure 4.

Binary image LBP image

Figure 4. Sample LBP image

4. EXPERIMENTAL RESULTS
Experiments are carried out on signature images chosen from local dataset and MCYT database. A total of 160
handwritten signatures (10 subjects of 16 each) are considered for performing experiments. In case of local dataset, for
each subject, out of 16 signatures, 8 signatures are used for training and 8 used are used for testing. From MCYT
standard database, 200 original and 100 forged signature images are considered. The training and test images include
both guanine and forged signatures of individual subjects.

For images from the training set, the LBP features extracted, as described in feature extraction algorithm, and are
stored as knowledge base for use in KNN classification, where as in case of using SVM classifier, the extracted features
from train images are input to SVM module for generating support vectors, which form the knowledge base. For
signature verification, subject signature from test is subjected to feature extraction, and the extracted features are input
to classifier for verification. The verification results are described in terms of true acceptance rate (TAR), false
acceptance rate (FAR) and false rejection rate (FRR). The details of the results obtained are presented in Table 2
through Table 7. From the results it is evident that both the classifiers yield verification results with FAR 0.0125, for
local database, and FAR of 0.1 and 0.15 for KNN and SVM, respectively, in case of MCYT database.

Volume 7, Issue 5, May 2019 Page 19


IPASJ International Journal of Information Technology (IIJIT)
Web Site: http://www.ipasj.org/IIJIT/IIJIT.htm
A Publisher for Research Motivation ........ Email:editoriijit@ipasj.org
Volume 7, Issue 5, May 2019 ISSN 2321-5976

Table 2. Results using KNN and SVM classifier for local database
Verification results
No of
Subjects KNN
Train/Test SVM
k=1 k=3
1 8/8 8 7 8
2 8/8 8 7 8
3 8/8 7 4 7
4 8/8 8 6 8
5 8/8 8 8 8
6 8/8 8 7 8
7 8/8 8 7 8
8 8/8 8 5 8
9 8/8 8 8 8
10 8/8 8 6 8

Table 3. Accuracy and FAR value for local database

Table 4. KNN classification For K=1

Table 5. KNN classification For K=3

Table 6. SVM classification

Table 7. Accuracy and FAR value of classifier for MCYT

Volume 7, Issue 5, May 2019 Page 20


IPASJ International Journal of Information Technology (IIJIT)
Web Site: http://www.ipasj.org/IIJIT/IIJIT.htm
A Publisher for Research Motivation ........ Email:editoriijit@ipasj.org
Volume 7, Issue 5, May 2019 ISSN 2321-5976

5. CONCLUSION
In this paper, an efficient approach for signature verification/authentication is proposed based upon the features
extracted using LBP operator. Experiments are performed using K-NN and SVM classifiers. It is observed that,
performance of KNN classifier is better over SVM classifier for MCYT database whereas the performance of both the
classifiers for signatures from local database is same. As future work, we are comparing the results obtained using basic
implementation of LBP with that of variants of LBP in literature by performing experiments on local dataset of
signature images.

References
[1] Prabhakar, S., Kittler, J.,Maltoni, D., O’Gorman, L., Tan, T.: Introduction to the special issue on biometrics:
progress and directions. IEEETran. Pattern Anal.Mach. Intell. 29(4), 513–516 (2007), Jain, A. K., Ross, A.,
Pankanti, S.: Biometrics: a tool for information security. IEEE Trans. Inf. Forensics Secur. 1(2), 125–143 (2006).
[2] R. Plamondon, “The design of an on-line signature verification system”, Theory to practice, International journal
of Pattern recognition and Artificial Intelligence, (1994)795-811.
[3] Batista, L., Rivard, D., Sabourin, R., Granger, E., Maupin, P.: State of the art in off-line signature verification.
In: Verma, B., Blumenstein, M. (eds.) Pattern Recognition Technologies and Applications: Recent Advances, pp.
39–62 (2008).
[4] Impedovo, D., Pirlo, G.: Automatic signature verification: the state of the art. IEEE Trans. Syst. Man. Cybern. C
Appl. Rev. 38(5), 609–635, 2008.
[5] Luiz G. Hafemann , Robert Sabourin and Luiz S. Oliveira, Offline Handwritten Signature Verification -
Literature Review, IEEE Explorer, ISSN: 2154-512X, 2017.
[6] Shashi Kumar et. al. “Off-line Signature Verification Based on Fusion of Grid and Global Features Using Neural
Networks”/ International Journal of Engineering Science and Technology Vol. 2(12), 2010, 7035-7044.
[7] M.S. Shirdhonkar et.al. “Off-Line Handwritten Signature Identification Using Rotated Complex Wavelet Filters”
IJCSI International Journal of Computer Science Issues, Vol. 8, Issue 1, January 2011 ISSN (Online): 1694-
0814www.IJCSI.org 478.
[8] Supinder Singh, and Amandeep Kaur, “Off-Line Signature Verification using Sub uniform Local Binary Patterns
and support vector machine “,Int'l Conf. on Chemical Engineering & Advanced Computational Technologies
(ICCEACT’2014) Nov. 24-25, 2014 Pretoria (South Africa).
[9] P. Vickram, Dr. A. Sri Krishna and D.Swapna ,Offline Signature Verification using Support Local Binary
Pattern, International Journal of Artificial Intelligence and Applications (IJAIA), Vol. 7, No. 6, November 2016.
[10] J. Ortega-Garcia, J. Fierrez-Aguilar, D. Simon, J. Gonzalez, M. Faundez, V. Espinosa, A. Satue, I. Hernaez, J. J.
Igarza, C. Vivaracho, D. Escudero and Q. I. Moro, "MCYT baseline corpus: A bimodal biometric database", IEE
Proceedings Vision, Image and Signal Processing, Special Issue on Biometrics on the Internet, Vol. 150, n. 6, pp.
395-401, December 2003.
[11] T. Ojala, M. Pietikäinen, and T. Maenpaa, “Multiresolution gray-scale and rotation invariant texture
classification with local binary patterns,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 24, no. 7,
pp. 971–987, 2002.
[12] Z. Guo, L. Zhang, and D. Zhang, “A completed modeling of local binary pattern operator for texture
classification,” IEEE Transactions on Image Processing (TIP), vol. 19, no. 6.
[13] Lowe, David G. (1999). "Object recognition from local scale-invariant features" (PDF). Proceedings of the
International Conference on Computer Vision. 2. pp. 1150–1157. doi:10.1109/ICCV.1999.790410.
[14] Lowe, David G. (2004). "Distinctive Image Features from Scale-Invariant Keypoints". International Journal of
Computer Vision. 60 (2): 91–110. CiteSeerX 10.1.1.73.2924 pp. 1657-1663, 2010.

Volume 7, Issue 5, May 2019 Page 21

You might also like