2023 IEEE 8th International Conference for Convergence in Technology (I2CT)

Pune, India. Apr 7-9, 2023

Offline Handwritten Signature Verification using

Image Processing Techniques
Dr. V. Hindumathi Jennifer Chalichemala Endla Ameya
2023 IEEE 8th International Conference for Convergence in Technology (I2CT) | 979-8-3503-3401-2/23/$31.00 ©2023 IEEE | DOI: 10.1109/I2CT57861.2023.10126203

Assosiate Professor in Electronics and Undergraduate in Electronics and Undergraduate in Electronics and
Communication Engineering Communication Engineering Communication Engineering
BVRIT HYDERABAD College of BVRIT HYDERABAD College of BVRIT HYDERABAD College of
Engineering for Women Engineering for Women Engineering for Women
Hyderabad, India Hyderabad, India Hyderabad, India
[email protected] [email protected] [email protected]

Vatadi Kavya Sri Bhople Vaibhavi

Undergraduate in Electronics and Undergraduate in Electronics and
Communication Engineering Communication Engineering
BVRIT HYDERABAD College of BVRIT HYDERABAD College of
Engineering for Women Engineering for Women
Hyderabad, India Hyderabad, India
[email protected] [email protected]

Abstract— The use of signatures for personal identification problem. When combined, these all lead to significant intra-
and verification is quite common. Signatures are validated for personal diversity. It is necessary to create a reliable system
many documents such as Bank cheques and legal transactions. that can not only take these considerations into account but
The necessity for effective automated solutions for signature also identify distinct forgeries.[2],[3],[21],[22][23]. Simple
verification has grown as signatures are now a prerequisite for forgeries, also referred to as unskilled forgeries, include a
both authorization and authentication in legal activities. Two forger who has access to the user's name but just not their
images—the original signature and the test signature—are signature. The fake signature in this instance resembles the
used as input in this process. Prior to feature extraction, these real signature in terms of shape and look. In the instances of
photos are pre-processed. To determine whether the signature
skilled counterfeiting, the forger is well conversant with the
is fake or not, the characteristics that were extracted are
compared, and the difference in error values between them is
customer's name and signature and is skilled at duplicating
examined. The key advantages of signature verification the signature. This fake signature appears just like the real
systems (SVS) over alternative verification technologies include one, making it incredibly challenging to spot.[16], [24], [25],
time and energy savings, a reduction in the risk of fraud [26]. Based on the fact that signatures cannot be recognised
during authentication, and a reduction in the likelihood of just by visual examination, automatic signature verification
human error during the signature process. is necessary. Although not flawless, automatic inspection is
extremely dependable and effective.[4], [5],[27],[28]. In this
Keywords— Handwritten Signature Verification, Feature paper, we examine the "consistency" of frequently employed
Extraction, Image Processing features in offline signature verification and define a
consistency model. We take into account the attributes that
I. INTRODUCTION were extracted from the input photos and produce an error
percentage that aids in determining if the signature is real or
For a long time, signatures served as the de facto method fake. In the section that follows, we outline the strategy we
of identification. A signature is essentially a written used to create the prototype system for online signature
fingerprint. Thus, the use of signatures in document verification. This paper is categorized into six sections, with
verification is essential. Perhaps the oldest and most the literature review appearing in the second section, the
extensively used method of confirming a person's methodology in the third section, the implementation in the
identification, particularly in financial, business and legal fourth section, the findings in the fifth section, and the
contexts, is the use of a signature. Signature characteristics conclusion in the sixth section.
are divided into the following categories: static and dynamic.
In essence, qualities that are readily apparent, such as shape,
are known as static qualities. It is possible for an experienced II. LITERATURE SURVEY
forger to replicate the static properties because they are A signature is an example of handwriting that has unique
readily observable. The features that are concealed, for characteristics and twists. It is possible for many signatures
example, the amount of pressure used at the tip of the pen, as to be unclear. The idea of signature verification is not new.
well as the placement of the pen tip, are known as dynamic More than 30 years of research have gone into it. Manual
qualities [1], [19], [20]. Due to high fashionable and signature verification was initially used, which is a poor
distinctive writing styles, it is tough to divide signature solution for this purpose. Since the invention of computers in
strokes, which adds to the challenge. The issue is made the 20th century, specialists have worked to develop new
worse by the non-repetitive character of the signature techniques for more precise signature verification.
variations caused by age, disease, location, and perhaps to a Information technology has been assisting with signature
degree, the individual’s metal state, accentuating the verification since 1960. [12] [13],[29]. A signature can be

979-8-3503-3401-2/23/$31.00 ©2023 IEEE 1

Authorized licensed use limited to: Nagarjuna College of Engineering & Technology. Downloaded on June 09,2025 at 04:50:01 UTC from IEEE Xplore. Restrictions apply.
 taken care of as a picture, and can be verified as a genuine or order to retrieve precise attributes in the subsequent stages.
counterfeit signature. Nagel and Rosenfeld presented the The pre-processing stage raises the image's quality and
initial approach for offline signature validation in the IEEE qualifies it for feature extraction. The prepossessing stage
Conference and Transaction in 1973. [14]. In order to draw involves converting an RGB image to grayscale and then
out information about pressure from grayscale photographs, into binary, centre of mass, normalisation, Resizing (which
Yeung, et al, devised a feature extraction method which is reduces the size of the signature images to the standard size
pseudo-dynamic in nature, in 1986.[15]. Research and of 512*512) as illustrated in Figure 1, Image cropping to
alternative Signature verification implementations have been capture the Region of Interest (ROI)
explored throughout the past ten years, concurrent with
technological improvement. Y. Inan and B. Sekeroglu
proposed [6] a method that aims to teach a neural network to
distinguish between the signatures of 27 individuals The
study discusses offline signature recognition using 27
different people's signatures to train a backpropagation
neural network. The signature is photographed and then pre-
processed by converting it into grayscale and resized to
50x50 pixels. The three categories of noise— Gaussian, salt
and pepper and Poisson—are incorporated to the image prior
to sending it the neural network training model. To train the Fig. 1. Resized Input Image
network, more than three hundred images with various
amounts of noise are employed. This is one particular area Feature Extraction: The effectiveness of the
where the paper's methodology falls short. The noise serves classification and the computing efficiency are greatly
as a crude mimic of the real environment. R. Plamondon and enhanced by feature extraction. In signature verification
S.N Srihari [7] conducted a survey on the nature of systems(SVS), it is vital to select a robust set of features. A
handwriting, the paper talks about Handwriting Interpretation feature vector is made using the features that were extracted
and identification. This paper's main goal is to provide a during this stage. Slope curve, Maximum horizontal and
thorough analysis of current approaches to automatic vertical Projection, center of Mass for horizontal and vertical
handwriting recognition. It details numerous recent Images(This gives the centroid), and Negative Selection
developments and shifts that have taken place in this area, Algorithm (NSA) are used to extract these features.
notably during the past ten years. A histogram of directed
gradients is used in the offline signature recognition IV. IMPLEMENTATION
technique that Patil, Almeida Bryan, Niketa Chettiar, and We present a system that requires two images as input as
Joyal Babu, proposed [8]. This paper’s aim was to automate illustrated in figure 2. For easier feature extraction and
the verification process of a signature because the visual greater enhancement, the input images are pre-processed. In
examination method now used and embraced by most the prepossessing phase, the images are converted from RGB
companies (such as banks) is laborious, time-consuming, and to Gray and then into binary images. After the conversion,
potentially unreliable for certifying documents like checks. A the images undergo the process of thinning. Thinning of
feedforward artificial neural network and an oriented images helps to narrow the strokes, present in the images to
gradient histogram are used to extract features from the achieve uniformity. Black Pixels present in the picture are
challenge. With four signature samples utilised for training stored in vectors. The vectors contain the black pixels’
individual users, the system's accuracy was 96.87%. positions, which is used to find the centre of mass. The
centre of mass is used to move the signature to the origin.
III. METHODOLOGY After this, the extra white spaces around the region of
The process of verifying a handwritten signature is a interest (that is, the signature) is removed thereby cropping
challenging one that must be approached from various angles the image. This cropped image is copied into a matrix which
and resolved through a number of phases. Many strategies then undergoes feature extraction. Feature extraction here
have lately been put out, and extensive research has been implies, Negative Selection Algorithm, aspect ratio, max
done on feature extraction and classification using Hidden horizontal and vertical projection, centre of mass and slope
Markov Model (HMM), Support Vector Machine (SVM), curve which are explained below. The features from the
Fast Fourier Transform (FFT), Multilayer Perceptron (MLP) input images are extracted as follows:
wavelets, and Neural networks(NN). [9][10]. Our proposed Negative selection Algorithm(NSA): Forrest and AI
Methodology is accomplished using MATLAB Commands presented Negative Selection Algorithm as a supervised
and functions [17]. It is displayed in the subsequent steps: learning strategy for computer security, network security,
Signature Dataset: We can utilise a dataset that will and anomalies detection issues. For detection and
enable us to attain the appropriate scalability based on the classification in images, negative selection algorithms are
system’s ability to scale. A collection of Signatures is employed.[11], [30], [31].
acquired from 15 individuals NSA = Total number of black pixels
Input signature images: Handwritten signatures (both the Total number of pixels
original and the test signature) are provided as input in the
form of images. The pre-processing of these images allows Aspect ratio: It is the proportion between the signature
for additional categorisation. image's height and width. This is done because a person's
signature might vary in breadth or height yet maintain a
Pre-process the signature images: The signature images
are subjected to a variety of pre-processing procedures in

2
Authorized licensed use limited to: Nagarjuna College of Engineering & Technology. Downloaded on June 09,2025 at 04:50:01 UTC from IEEE Xplore. Restrictions apply.
 roughly similar ratio. To determine aspect Ratio, the formula After being extracted from the corresponding image, the
stated below is applied: features are saved into two vectors. An error value is
produced after comparing these vectors. The error value
Aspect Ratio = Height of the image shows how well the two images match up. Consistency here,
Width of the image refers to the static characteristics of the signature, such as the
font, strokes, spacing between the letters, and the width and
Maximum horizontal and vertical Projection: The height of the individual letters. The tested signature is
Horizontal histogram is established by determining the classified as counterfeit if the resulting error value exceeds a
amount if black pixels present within every row of the input predetermined threshold error value; if it is less than the
images.The maximum horizontal histogram is the row that threshold value, it is classified as genuine.
has the most black pixels. Similar to the horizontal
histogram, a vertical histogram is computed by inspecting
every column of the signature image and identifying the one V. RESULTS
with the greatest amount of black pixels. A total of 75 signatures were employed to produce the
research's findings. There are three samples for real
Centre of Mass : Split the signature images in two equal
signatures and two samples for counterfeit signatures for a
parts and find the centre of mass for individual parts. The
set of fifteen individuals. The results of two different test
centre of mass is calculated both horizontally and vertically,
situations are displayed in Figures 3 and 5. The graphical
that is, along the x-axis and y-axis.
depiction of the error and static features of the signatures is
Slope Curve: According to the font and strokes in the shown in Figures 4 and 6.
signature the slope curve is calculated
Test Case 1: Genuine Signature
Two Input Images are taken as illustrated in Fig. 3 The
first Image is the Pre-saved signature of person 1 (original
Signature) and the second image is the test image done by
person 1. Fig. 4 illustrates the graphical representation of
error between the two images.

Fig. 3. Case 1.

Test Case 2: Forged Signature

Two Input Images are taken as shown in Fig. 5 The first
Image is the Pre-saved signature of person 1 (original
Signature) and the second image is the test image, containing
the same signature but is handwritten by a different person.
Fig. 6 demonstrates the graphical representation of error
between the two images.

Fig. 2. Flow Chart of our offline signature Verification Method Using

MATLAB
Fig. 4. Error Graph of case 1.

3
Authorized licensed use limited to: Nagarjuna College of Engineering & Technology. Downloaded on June 09,2025 at 04:50:01 UTC from IEEE Xplore. Restrictions apply.
 [7] R. Plamondon and S. N. Srihari, “Online and off-line handwriting
recognition: a comprehensive survey,” IEEE Transactions on Pattern
Analysis and Machine Intelligence, vol. 22, no. 1, pp. 63–84, 2000,
doi: 10.1109/34.824821
[8] Patil, P., Almeida, B., Chettiar, N., & Babu, J. (2017, December).
Offline signature recognition system using histogram of oriented
gradients. In 2017 International Conference on Advances in
Computing, Communication and Control (ICAC3) (pp. 1-5). IEEE.
[9] P. G and Dr. A. K.S, “OFFLINE HANDWRITTEN SIGNATURE
RECOGNITION USING EDGE HINGE AND EDGE
EXTRACTION TECHNIQUES AND MANHATTAN DISTANCE
CLASSIFIER,”, International Journal of Electronics and
Communication Engineering anf Technology, Vol 1, Issue 8, Feb.
Fig. 5. Case 2. 2017.
[10] D. Morocho, J. Hernandez-Ortega, and A. Morales, “On the
Evaluation of Human Ratings for Signature Recognition,”,
International Carnahan Conference on Security Technology, 2016.
[11] E. Bendiab and M. K. Kholladi, “The Negative Selection Algorithm:
a Supervised Learning Approach for Skin Detection and
Classification.,”, IJCSNS International Journal of Computer Science
and Network Security, VOL.10 No.11, Nov. 2010.
[12] N. Sharma, S. Gupta, and P. Mehta, “A Comprehensive Study on
Offline Signature Verification,” Journal of Physics: Conference
Series, vol. 1969, no. 1, p. 012044, Jul. 2021,
[13] Pal, S., Alaei, A., Pal, U., & Blumenstein, M. (2016, April).
Performance of an off-line signature verification method based on
texture features on a large indic-script signature dataset. In 2016 12th
IAPR workshop on document analysis systems (DAS) (pp. 72-77).
IEEE.
[14] Nagel, R. N., & Rosenfeld, A. (1973, October). Steps toward
handwritten signature verification. In Proc. First Int. Joint Conf. on
Pattern Recognition, Washington, DC (pp. 59-66).
[15] Yeung, D. Y., Chang, H., Xiong, Y., George, S., Kashi, R.,
Fig. 6. Error Graph of Case 2. Matsumoto, T., & Rigoll, G. (2004, July). SVC2004: First
international signature verification competition. In International
conference on biometric authentication (pp. 16-22). Springer, Berlin,
VI. CONCLUSION Heidelberg.
Offline signature verification is of dire need as it is [16] Soleimani, A., Araabi, B. N., & Fouladi, K. (2016). Deep multitask
metric learning for offline signature verification. Pattern Recognition
required and applicable in many fields of work such as, Letters, 80, 84-90.
Finance, legal transactions, business, retail and many more [17] D. Engin, A. Kantarcı, S. Arslan, and H. Ekenel, “Offline Signature
[18]. Consequently, a system that can distinguish between Verification on Real-World Documents.” Accessed: Nov. 27, 2022.
authentic and fake signatures is required. The working [18] Zimmerman, T. G., Russell, G. F., Heilper, A., Smith, B. A., Hu, J.,
implementation of offline signature verification system that Markman, D., ... & Drews, C. (2004, August). Retail applications of
can determine whether a signature is real or fake is provided signature verification. In Biometric Technology for Human
in this paper. Our test outcomes demonstrate that even with a Identification (Vol. 5404, pp. 206-214). SPIE.
lesser number of signature samples from each individual, the [19] Tahir, N. M., Ausat, A. N., Bature, U. I., Abubakar, K. A., & Gambo,
I. (2021). Off-line Handwritten Signature Verification System:
system is able to validate the signature by taking features like Artificial Neural Network Approach. Int. J. Intell. Syst. Appl, 13, 45-
size, height, width, strokes etc, into consideration. 57.
[20] Iranmanesh, V., Ahmad, S. M. S., Adnan, W. A. W., Malallah, F. L.,
REFERENCES & Yussof, S. (2013, December). Online signature verification using
neural network and pearson correlation features. In 2013 IEEE
[1] S. K. Ahmed, A. K. Ramasamy, A. S. Mohd. Khairuddin, and J. conference on open systems (ICOS) (pp. 18-21). IEEE.
Omar, “Automatic online signature verification: A prototype using
neural networks,” IEEE Xplore, Jan. 01, 2009. [21] Huang, K., & Yan, H. (2002). Off-line signature verification using
structural feature correspondence. Pattern Recognition, 35(11), 2467-
[2] Y. Qi and B. R. Hunt, “Signature verification using global and grid 2477. D. Morocho, J. Hernandez-Ortega, and A. Morales, “On the
features,” Pattern Recognition, vol. 27, no. 12, pp. 1621–1629, Dec. Evaluation of Human Ratings for Signature Recognition,”,
1994, doi: 10.1016/0031-3203(94)90081-7. International Carnahan Conference on Security Technology, 2016.
[3] B. Kareem Abd, Q. Khaled Abood, and N. A.Z. Abdullah, “Hand [22] Samuel, D., & Samuel, I. (2010). Novel feature extraction technique
Written Signature Verification based on Geometric and Grid for off-line signature verification system. International Journal of
Features,” Jan. 2015.. Engineering Science and Technology, 2(7), 3137-3143
[4] F. Leclerc and R. Réjean, “Automatic Signature Verification: The [23] Ferrer, M. A., Vargas, J. F., Morales, A., & Ordonez, A. (2012).
State of the Art - 1989-1993.,” International Journal of Pattern Robustness of offline signature verification based on gray level
Recogmition and Artificial Intelligence, 8(3), Jun. 1994. features. IEEE Transactions on Information Forensics and
[5] R. Sabourin, R. Plamondon, and G. Lorette, “Off-line Identification Security, 7(3), 966-977.
With Handwritten Signature Images: Survey and Perspectives.” [24] Hafemann, L. G., Sabourin, R., & Oliveira, L. S. (2016, July).
Accessed: Nov. 25, 2022. Writer-independent feature learning for offline signature verification
[6] Y. Inan and B. Sekeroglu, “Signature Recognition Using using deep convolutional neural networks. In 2016 international joint
Backpropagation Neural Network,” 13th International Conference on conference on neural networks (IJCNN) (pp. 2576-2583). IEEE.
Theory and Application of Fuzzy Systems and Soft Computing — [25] Rantzsch, H., Yang, H., & Meinel, C. (2016, December). Signature
ICAFS-2018, pp. 256–261, Dec. 2018, doi: 10.1007/978-3-030- embedding: Writer independent offline signature verification with
04164-9_35 deep metric learning. In International symposium on visual
computing (pp. 616-625). Springer, Cham.

4
Authorized licensed use limited to: Nagarjuna College of Engineering & Technology. Downloaded on June 09,2025 at 04:50:01 UTC from IEEE Xplore. Restrictions apply.
 [26] Hafemann, L. G., Sabourin, R., & Oliveira, L. S. (2016, December). [29] Ferrer, M. A., Vargas, F., Travieso, C. M., & Alonso, J. B. (2010,
Analyzing features learned for offline signature verification using October). Signature verification using local directional pattern (LDP).
Deep CNNs. In 2016 23rd international conference on pattern In 44th Annual 2010 IEEE International Carnahan Conference on
recognition (ICPR) (pp. 2989-2994). IEEE. Security Technology (pp. 336-340). IEEE.
[27] Plamondon, R., & Lorette, G. (1989). Automatic signature [30] Elgammal, A., Muang, C., & Hu, D. (2009). Skin detection-a short
verification and writer identification—the state of the art. Pattern tutorial. Encyclopedia of Biometrics, 4, 1218-1224.
recognition, 22(2), 107-131. [31] Chen, W. C., & Wang, M. S. (2007). Region-based and content
[28] Impedovo, D., & Pirlo, G. (2008). Automatic signature verification: adaptive skin detection in color images. International journal of
The state of the art. IEEE Transactions on Systems, Man, and pattern recognition and artificial intelligence, 21(05), 831-853.
Cybernetics, Part C (Applications and Reviews), 38(5), 609-635.

5
Authorized licensed use limited to: Nagarjuna College of Engineering & Technology. Downloaded on June 09,2025 at 04:50:01 UTC from IEEE Xplore. Restrictions apply.