ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal
Impact Factor: 7.301 Volume 3, Issue 16, May 2023
Signature Verification using CNN
Hrishikesh Mhaske1, Rushikesh Bhalerao2, Sanket Walke3, Vaibhav Gholap4, Prof. Puja Lingampalli5
1,2,3,4
Student, Computer Engineering, Rajiv Gandhi College of Engineering, Karjule Harya, India
5
Assistant Professor, Computer Engineering, Rajiv Gandhi College of Engineering, Karjule Harya, India
Abstract: One of the most popular verification biometrics is the signature. The usage of handwritten
signatures incheques, applications, letters, forms, minutes, etc. A person's handwritten signature must be
individually identified because each individual's signature is unique by nature. Verifying signatures is a
popular utilised technique for verifying someone while they are away. Human verification can be
inaccurate and occasionally [Link] most common method for confirming a person or a private is with
a signature. A person's signature is used to identify them in all social, professional, and commercial
contexts. The word "signature verification" is extremely important because it could be misused and lead to
significant losses. The signature may be a behavioural biometric trait that combines the signer's
neuromotor characteristics (e.g., how our brain and muscles, among other things, shape how we tend to
sign) as well as sociocultural influences (e.g., the differences between Western and Asian styles). Through
the ages, United Nations agency experts have constructed signature examinations to verify the validity of
sample-supported rhetorical analysis.
Keywords: CNN (Convolutional Neural Network),Signature Verification, Support Vector Machine,
Biometric Analysis.
I. INTRODUCTION
We intended to develop the deep learning technology to recognise the human handwriting in a way that would also be
simple for someone with less technical experience to utilise. The suggested technique is designed to determine whether
a human handwritten signature is real or fake. A greater variety of photos, including examples of both real and fake
signatures, are gathered as samples. For each class of photographs that were categorised as input images, a different
number of images were collected. To stop hostile individuals from using fake signatures, we suggested a Deep Learning
(DL) based offline signature verification solution. The Convolutional Neural Network (CNN) is the Deep Learning
(DL) technique employed in the study. If more feature extraction techniques are added to assist the CNN method and
the human hand signature is correctly classified, it is projected that the success of the results will grow.
The method for verifying human signatures uses a deep learning algorithm, which has numerous "layers" of neural
network algorithms. As signatures pass through each layer, a condensed version of the data is presented to the following
layer. The majority of machine learning techniques are effective when applied to datasets of up to a few hundred
features. The quantity of features in an unstructured dataset like a picture, however, makes this approach difficult or
impossible. As the deep learning algorithm traverses each layer of the neural network, it gradually gains more
knowledge about the image of the signatures. The final output layer contains the results of the signature, which are then
shown on the screen as either a real or fake signature.
II. LITERATURE SURVEY
The area of Handwritten Signature Verification has been broad lyre search edin the last decades and still remains as an
open research problem. This project focuses on offline signature verification, characterized by the usage of
static(scanned)images of signatures, where the objective is to discriminate if a given signature is genuine.
1. BIOMETRIC SIGNATURE VERIFICATION SYSTEM
The phrase "biometric signature verification system based on freeman chain code and k-nearest neighbour" [1],
Three issues were found during the analysis of the problem's historical context in Stage The first one is relevant to
the SVS as a whole. Some solutions to this issue are defined since signatures are a sort of biometric that can alter
Copyright to IJARSCT DOI: 10.48175/568 376
[Link]
ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal
Impact Factor: 7.301 Volume 3, Issue 16, May 2023
with mood, environment, and age. A decent signature database needs to be updated at certain intervals in order to
remain current and useful. In addition, one must sign consistently in order to create a string of signatures that are
very similar to one another. The second issue is with the FCC generation, which was unable to extract from the
damaged portions of the signature. the greatest contiguous portion of The signature is picked in order to reveal the
FCC. The third issue had to do with verification in order to have a decent outcome. To get the desired results using
k-NN, earlier steps, particularly pre-processing and feature extraction, must function effectively.
2. MULTIPLE NEURAL CLASSIFIERS
The phrase "Signature verification using multiple neural classifiers" [2] - On Sun's Spark System, a prototype
recognition system was developed using C. Ten separate people's samples were obtained for the experiment. Each
person provided fifteen authentic signature samples, which were then collected. In addition, 100 random forgeries
were utilised to evaluate the system. A series of trials were run to gauge the approach's effectiveness. In every
experiment, five randomly chosen samples of each person's actual signature were used to train the classification
nets. These nets had the same number of output nodes as there were participants in the experiment.
3. TWO-STAGE NEURAL NETWORK CLASSIFIER
"A new neural network classifier-based two-stage neural network signature verification technique" [3] - This
research suggests a fresh method for off-line signature verification and identification. The complete system is built
on a two-stage neural network classifier with one class per network and 160 features divided into three subsets.
Only small, fixed-size neural networks need to be trained during the first stage's training procedure, but the
training method for the second stage is simple. The majority of our design work went towards incorporating the
majority of the intelligence into the system's structure. There was no feature reduction method employed, and the
guiding principle for choosing which features to include and which to exclude was to "use all features." and let the
neural networks decide which ones are significant and which ones are not"
4. SIGNATURE IMAGE GENERATION
"Synthetic off-line signature image generation" [5] In this work, a brand-new mechanism for creating static,
offline signatures of new identities is proposed. In particular, the random variables of a statistical distribution of
the features of the global signature are used to create the signature of the new synthetic identity. The outcomes
closely resemble actual signature forms and characteristics of writing styles that are inferred from static signature
databases. With the introduction of a natural variability from the synthetic individual attributes, new instances and
forgeries of the synthetic identities are obtained. An ink deposition model based on a ballpoint is created for the
creation of realistic static signature images as an added novelty. The static signature generator's range has been
established to match the performance of two public databases as well as that of synthetic databases.
5. STATIC SIGNATURE SYNTHESIS
"Static signature synthesis: A biometrics method inspired by neuromotor behaviour" [6], we provide a novel
technique in this research for creating fake handwritten signature images for biometric applications. The methods
we present mimic the motor equivalence mechanism, which splits human handwriting into two stages: the
formulation of an effector independent action plan and its execution via the associated neuromuscular pathway.
On a spatial grid, the activity plan is shown as a trajectory. This includes the signature text as well as its flourish,
if any. Applying a kinematic Kaiser filter on the trajectory plan simulates the neuromuscular path. The pen speed,
which is produced using a scalar version of the sigma lognormal model, determines the length of the filter. A
deposit of ink Using a model to simulate the pen trajectory pixel by pixel, realistic static signature images are
produced.
The range of the synthesis parameters as well as the lexical and morphological characteristics of the synthesised
signatures have been determined using real signature databases like the GPDS960 and MCYT Off-line Grey
Signature corpuses. The results of the performance experiments demonstrate that it is possible to generate
synthetic identities with varying levels of stability and skilled forgers. As a result, datasets of artificial signatures
Copyright to IJARSCT DOI: 10.48175/568 377
[Link]
ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal
Impact Factor: 7.301 Volume 3, Issue 16, May 2023
can be produced that perform similarly to databases of actual signatures. Additionally, depending on the demands
of the researcher, we can modify the generated dataset to produce sophisticated forgeries or straightforward
forgeries that are simpler to spot. The average level of confusion between actual and artificial signatures according
to perceptual evaluation is 44.06%, demonstrating the artificial signatures' realism. By examining how the type of
pen and user count affect an automatic signature validator, the usefulness of the synthesised signatures is
illustrated.
6. MODULES
Add the specified image to the dataset and CNN training manual Utilising the keraspreprocessing image data
generator tool, we must input our data set and construct size, rescale, range, zoom range, and horizontal flip. Then,
using the data generator tool, we import our image dataset from the folder.
FIG - 1. PHASES OF HUMAN SIGNATURE VERIFICATION
In addition to setting the target size, batch size, and class-mode from this function, we also set the train, test, and
validation parameters here. The module will be trained using AlexNetWe create training steps for each epoch, total
epochs, an image is inputted into a convolutional neural network (ConvNet/CNN), a deep learning algorithm, which
assigns emphasis (learnable weights and biases) to distinct aspects in the image of the signature and be able to
distinguish between different signatures.
Fig. 2. Layers in Convolution Neural Network
Copyright to IJARSCT DOI: 10.48175/568 378
[Link]
ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal
Impact Factor: 7.301 Volume 3, Issue 16, May 2023
Using the Django framework to deploy the model and forecast the outcome. The trained deep learning model is
transformed into a hierarchical data format file in this module, which is then used by our Django framework to improve
the user experience and forecast if a particular signature is genuine or fake.
7. IMPLEMENTATION AND RESULT ANALYSIS
7.1 Data Acquisition
To develop a knowledge base for each and every person, handwritten signatures are gathered, and some of the
signatures' distinctive characteristics are retrieved. For assessing the effectiveness of the signature verification
system and contrasting the outcomes achieved using the other techniques on the same database, a standard
database of each person's authentic and forgeried signatures is required.
Fig. 3. Genuine and Forged Signatures
7.2 Pre-Processing
The scans of each signature are done in grey. This stage's goal is to standardise and prepare the signature for
feature extraction. The pre-processing procedure enhances the signatures' quality and makes them better suited for
feature extraction. A grayscale signature image is included in the prepossessing stage and converted to binary to
facilitate feature extraction. The signatures received by signature are obtained in various sizes, thus resizing is
done to make them standard size, which will make the signatures to standard size 256*256.
Fig. 4. Pre-Processing of Signatures
Copyright to IJARSCT DOI: 10.48175/568 379
[Link]
ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal
Impact Factor: 7.301 Volume 3, Issue 16, May 2023
7.3 Classification
Every picture in the signature goes through a sequence of alternating convolution and max pooling layers. A
predetermined number of feature maps are generated after each image in the signature has undergone the
convolution process. These feature maps are then fed into the max pooling layer, which then generates pooled
feature maps using the feature maps acquired from the convolution layer put before it. The following convolution
layer receives this pooled feature map, and so on until the fourth maximum pooling layer is reached. The last max
pooling layer's pooled feature map is received, flattened, and delivered into the fully linked layers. The model will
be finalised after multiple iterations of forward and backward propagation. After the model has been trained, it
will be possible to forecast whether a signature is genuine or fake.
7.4 Overall Result
The database containing the signatures is tested after training the CNN models on the images of the signatures,
and the outcome indicates whether the matching signature is real or fake.
Fig. 5. Sample output generated
III. CONCLUSION AND FUTURE WORK
Embedded devices can use Convolutional Neural Networks, a very powerful and effective technique. The effectiveness
of the algorithm can be confirmed using the aforementioned tests. All of these tests yield very comparable results. The
training of signature datasets obtained from multiple perspectives is a crucial parameter to take into account, according
to algorithm testing. The public will be helped in using their signatures in the safest and most effective way possible
with the help of this intelligent human signature verification technology.
The project's future development will involve integrating online apps with artificial intelligence to use the human
hand signature verification technique.
IV. ACKNOWLEDGMENT
The heading of the Acknowledgment section and the References section must not be numbered. Causal Productions
wishes to acknowledge Michael Shell and other contributors for developing and maintaining the IEEE LaTeX style files
which have been used in the preparation of this template.
REFERENCES
[1] Bharadi, V. A. & Singh, V. I. (2014), `Hybrid Wavelets based Feature Vector Generation from Multidimensional
Data set for On-line Handwritten Signature Recognition', 5th International Conference- Conuencethe Next
Generation Information Technology Summit (Conuence pp. 561-568).
[2] Chang, H., Dai, D.,Wang, P. & Xu, Y. (2012), `Online Signature Verification Using Wavelet Transform of
Feature Function Architecture of an Online Signature Verification System', 11(2011), 3135-3142.
Copyright to IJARSCT DOI: 10.48175/568 380
[Link]
ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal
Impact Factor: 7.301 Volume 3, Issue 16, May 2023
[3] Fernandes, J. & Bhandarkar, N. (n.d.), `Enhanced online signature verification system', International Journal of
Emerging Trends and Technology in Computer Science (IJETTCS), Volume 3, Issue 6, November - December
2014, pp. 205-209, ISSN 2278-6856.
[4] Fierrez-aguilar, J., Krawczyk, S., Ortega-garcia, J. & Jain, A. K.(2005), `Fusion of Local and Regional
Approaches for On-Line Signature Verification', Iwbrs 2005 LNCS 3781, 188-196.
[5] Hafemann, L. G., Sabourin, R. & Oliveira, L. S. (2017), `Learning features for online handwritten signature
verification using deep convolutional neural networks', Pattern Recognition 70, 163-176.
[6] Iranmanesh, V., Ahmad, S. M. S., Adnan, W. A. W., Yussof, S., Arigbabu, O. A. &Malallah, F. L. (2014), `Online
Handwritten Signature Verification Using Neural Network Classifier Based on Principal Component Analysis',
The Scientific World Journal 2014, 1-9.
[7] Jain, A. K., Ross, A. A. & Nandakumar, K. (2011), Introduction, in `Introduction to Biometrics', Springer, pp. 1-
49.
[8] Kaur, M. R. & Choudhary, M. P. (2015), `Handwritten Signature Verification Based on Surf Features Using
Hmm', 3(1), 187-195.
[9] Khalil, M., Moustafa, M. & Abbas, H. (2009), `Enhanced DTW based on-line signature verification', Image
Processing (ICIP), 2009 16thIEEE International Conference on pp. 2713-2716.
[10] Liu, Y., Yang, Z. & Yang, L. (2015), `Online signature verification based on dct and sparse representation', IEEE
transactions on cybernetics 45(11), 2498-2511.
[11] Nagbhidkar, K. P. &Bagdi, P. V. (2015), ` Online Signature Verification on smart phone using discrete wavelet
transforms.', 2(2), 1-6.
[12] Nanni, L., Maiorana, E., Lumini, A. &Campisi, P. (2010), `Combining local, regional and global matchers for a
template protected on-line signature verification system', Expert Systems with Applications 37(5), 3676-
[Link]: [Link]
[13] Parodi, M. & Gomez, J. C. (2014), `Legendre polynomials-based feature extraction for online signature
verification. Consistency analysis of feature combinations', Pattern Recognition 47(1), [Link]:
[Link]
[14] Plamondon, R., Pirlo, G. &Impedovo, D. (2014), Online signature verification, in `Handbook of Document Image
Processing and Recognition', Springer, pp. 917-947.
[15] Plotz, T. & Fink, G. a. (2009), `Markov models for offline handwriting recognition: A survey', International
Journal on Document Analysis and Recognition 12, 269-298.
[16] Rua, E. A. & Castro, J. L. A. (2012), `Online signature verification based on generative models', IEEE Trans.
Syst., Man, Cybern. B, Cybern 42(4), 1231-1242.
[17] Saffar, M. H., Fayyaz, M., Sabokrou, M. & Fathy, M. (2018), `Online signature verification using deep
representation: A new descriptor', arXiv preprint arXiv:1806.09986.
[18] Sharma, A. & Sundaram, S. (2016), `An enhanced contextual dtw based system for online signature verification
using vector quantization', Pattern Recognition Letters 84, 22-28.
[19] Thumwarin, P., Pernwong, J. & Matsuura, T. (2013), `FIR signature verification system characterizing dynamics
of handwriting [Link]: [Link]
Copyright to IJARSCT DOI: 10.48175/568 381
[Link]