[Link].
org © 2022 IJCRT | Volume 10, Issue 5 May 2022 | ISSN: 2320-2882
Offline Signature Verification using Python
Nakshita Pramod Kinhikar Dr.K.N. Kasat
PG Scholar Asst Professor
Department of Electronics and Telecommunication
Prof Ram Meghe College of Engineering and Management,Badnera.
Abstract— Every person has his/her own unique signature that is used mainly for the purposes of
personal identification and verification of important documents or legal transactions. There are two kinds
of signature verification: static(offline) and dynamic(online). Static verification is the process of verifying
an electronic or document signature after it has been made. Offline signature verification is not efficient
and slow for a large number of documents. To overcome the drawbacks of offline signature verification,
we have seen a growth in online biometric personal verification such as fingerprints, eye scan etc.
In this project , offline signature verification using Convolutional Neural Network (CNN) is proposed.
CNN is a type of neural network model which allows us to extract higher representations for the image
content. CNN takes the image's raw pixel data, trains the model, then extracts the features automatically
for better classification . The main advantage of CNN compared to its predecessors is that it automatically
detects the important features without any human supervision also it has the highest accuracy among all
algorithms that predicts images.
Keywords— Offline signature, Image processing, Convolutional Neural Network , Artificial Neural
Network , Authentication, Accuracy and Security.
I. INTRODUCTION
In a world which is progressing towards innovation, signature actually assumes the most crucial part in ID of
a specific individual. As years cruise by, instances of phony is likewise expanding in an incredible number.
Hence, signature check framework is request of an opportunity to further develop the confirmation
interaction and give secure means to approval of authoritative archives. The mark confirmation frameworks
help to separate between the first and phony marks.
This section presents data about advancements utilized in project that are Image Processing and
Convolutional Neural Network.
IJCRT2205942 International Journal of Creative Research Thoughts (IJCRT) [Link] i22
[Link] © 2022 IJCRT | Volume 10, Issue 5 May 2022 | ISSN: 2320-2882
1.1.1 Image Processing
Image processing is is a strategy to play out certain procedure on a picture, to get an improved picture or to
extricate some valuable data from it.
1.1.2 Artficial Neural Network
Artificial Neural Network is an endeavor to reproduce the organization of neurons that make up a human
cerebrum so the PC will actually want to learn things and settle on choices in a humanlike way.
1.1.3 Convolutional Neural Network
A convolutional neural organization (CNN) is a kind of counterfeit neural organization utilized in picture
acknowledgment and handling that is explicitly intended to deal with pixel information.
II. LITERATURE SURVEY
The survey paper we have studied is tabulated as follows:
Sr. N0 Ref. No, Author Work Done By Findings by Conclusion
and Year Author Author
1. KESANA implemented an Extensive performing
MOHANA efficient simulation texture features
LAKSHMI, framework for analysis extraction by
TUMMALA off line signature disclosed that the computing
RANGA BABU recognition proposed spatial
[1] system by methodology dependence
“AN utilizing NSCT achieved matrix and
EFFICIENT framework superior utilized NSCT
ALGORITHM performance over algorithm for
FOR HAND the conventional multi-scale
WRITTEN retrieval system decomposition
SIGNATURE by achieving and
RECOGNITION higher mAP and directionality.
USING mAR
TRANSFORM
BASED
APPROACH
WITH IMAGE
STATISTICS”
2. A. B. Jagtap and Both the The feature set These features
R. S. Hegadi envelopes are consists of set are coupled
[2] fused by features such as with support
“Offline performing union large and small vector machine
Handwritten operation and Eigen values classifier that
Signature their covariance computed from lead to 98.5% of
Recognition is computed upper envelope accuracy
Based on Upper and lower
and Lower envelope and its
Envelope Using union values
Eigen Values”
IJCRT2205942 International Journal of Creative Research Thoughts (IJCRT) [Link] i23
[Link] © 2022 IJCRT | Volume 10, Issue 5 May 2022 | ISSN: 2320-2882
3. D. Morocho , work explores Present a FDE analysis
A. Morales, human-assisted crowdsourcing give better
[Link] , and R. signature experiment to understanding of
VeraRodriguez recognition establish the output
[3] including a human
“Towards baseline baseline
human assisted performance of performance for
signature human signature
recognition: recognition of recognition tasks
Improving signatures and and a
biometric the analysis of novel attribute
systems through manual attribute- based semi-
attribute based based signature automatic
recognition” recognition. signature
verification
system inspired
in FDE analysis.
4. R. Sa-Ardship They has the adaptive These problems
and reviewed offline variance can be
K. Woraratpanya signature reduction considered for
[4] verification algorithm is improving the
“Offline schemes in their proposed to system
Handwritten paper. They have retain
Signature considered the
Recognition Artificial Neural
Using Polar- Network
Scale Technique
Normalization”
5. Vinayak Jadhav, paper presents a For For verification
Nikhil kadam, method of offline implementation of signatures
Paresh Keluskar, signature of above this some novel
Ayyaj Khan[5] verification using paper uses Feed features needs to
“Artificial artificial neural Forward Neural be extracted. The
Neural Network network Network (FFNN) extracted features
Based Signature approach using Error Back are used to train
Verification” propagation a neural network
algorithm for using error back
recognition and propagation
verification of training
signatures of algorithm
individuals
III. Implementation Methodology
1.1 Proposed System
In this project, we center around removing preprocessed information from the data set to prepare and test the
organization utilizing brain network methods to characterize a signature as certified or fabrication. The
acknowledgment and check of disconnected signature tests utilizing counterfeit brain network is pertinent as
it follows a worldview which models human learning designs.
IJCRT2205942 International Journal of Creative Research Thoughts (IJCRT) [Link] i24
[Link] © 2022 IJCRT | Volume 10, Issue 5 May 2022 | ISSN: 2320-2882
Data Acquisition/Signature Database
The signature database is gathered from MCYT-75 disconnected signature corpus data set. Every signature is
finished utilizing a WAMCOM Intuous inking pen. In which 15 certifiable and 15 imitation signature tests
are given for every one of 75 clients in data set. The imitation mark in the MCYT data set is the combination
of irregular, straightforward and gifted falsifications.
a) Training Stage - A training stage consist of following two steps
i. Retrieval of a signature image from a database.
ii. Neural network training.
b) Testing Stage - A testing stage consists of following two steps
i. Retrieval of a signature to be tested from a database.
ii. Checking output generated from a neural network.
The methodology begins by examining pictures into the PC utilizing fringe gadgets, then changing their
quality through picture upgrade, trailed by highlight extraction and brain network preparing, lastly checks
whether a signature is real or fake.
The general engineering of our unique acknowledgment framework follows: Signature securing,
Preprocessing, Feature extraction, and Classification. Offline signatures are made on papers. By and large in
any picture handling application preprocessing is expected to eliminate adjustment, from the first info
picture. Any normal scanner with sufficient resolution can be used as an image attainment device for offline
operation. Signatures are scanned in gray, using following equation as: (1) Gray colour = (0.299*Red) +
(0.5876*Green) + (0.114*Blue)
Scaling Let H be the height of the inputted image & W be the width of the inputted image. We can fit the
image uniform at 100*100 pixels by using the following equation as:
(2) Xnew = (Xold * 100)/H; Where Xnew&Xold are calculated & original X coordinate
(3) Ynew= (Ynew * 100)/W; Where Ynew&Yold are calculated & original Y coordinate.
Sound Reduction Images are defiled due to coming from translating blunders or boisterous channels. A
picture additionally gets corrupted on account of the impeding impacts because of light and different articles
IJCRT2205942 International Journal of Creative Research Thoughts (IJCRT) [Link] i25
[Link] © 2022 IJCRT | Volume 10, Issue 5 May 2022 | ISSN: 2320-2882
in the environment.
Background Elimination- We involved Thresholding strategy for recognize the signature from the
background. In proposed application, we are worried in dim items on light foundation and subsequently a
limit esteem T entitled as the brilliance edge is reasonably picked and applied to picture. After the
Thresholding, the pixels of the mark would be 1 and different pixels which have a place with the background
would be 0.
The brightness threshold can be chosen such that it satisfies the following conditions;
Suppose image pixels f(x, y) then, If f(x, y) ≥ T Then f(x, y) = Background
Else f(x,y) = Object
1.2 Architecture
IJCRT2205942 International Journal of Creative Research Thoughts (IJCRT) [Link] i26
[Link] © 2022 IJCRT | Volume 10, Issue 5 May 2022 | ISSN: 2320-2882
1.3 Working
In this project, we first train all the signature images. These trained and tested images are saved in the
training and testing folder. Then it gets the image name and path as input from the user and starts verification
on it step by step using Convolutional neural networks. In this process, first image preprocessing happens. In
image preprocessing binarization, segmentation is done. The extracted image is then used to train the
classifier, and if it matches, it produces a genuine output; otherwise, it produces a fraudulent output.
IV. OUTPUT
IJCRT2205942 International Journal of Creative Research Thoughts (IJCRT) [Link] i27
[Link] © 2022 IJCRT | Volume 10, Issue 5 May 2022 | ISSN: 2320-2882
V. CONCLUSION
In this paper, we presented different formulations for learning representations for offline signature
verification. Analyzing the above results shows that Convolutional Neural Networks are better for the
classification of signatures. The intuition to classify between genuine signatures and forgeries (regardless
of the user) by learning the visual cues has improved the accuracy. The significant improvement in the
accuracy is also due to the new architecture inspired by GoogleNet, which worked more widely than going
deeper. Hence, from the above experimental results, it is clear that the InceptionSVGNet Architecture is
more efficient in identifying patterns in images by using the wider networks. This pattern will continue for
future work, with researchers proceeding to investigate better feature sets (using Deep Learning networks)
and approaches to improve arrangement with a limited number of tests. Techniques based on ensembles of
classifiers, specifically methods for dynamic choice, are likewise encouraging in this field.
VI. REFERENCES
[1] Kesana Mohana Lakshmi and Tummala Ranga Babu, “ AN EFFICIENT ALGORITHM FOR HAND
WRITTEN SIGNATURE RECOGNITION USING
[2] A. B. Jagtap and R. S. Hegadi, “Offline Handwritten Signature Recognition Based on Upper and
Lower Envelope Using Eigen Values,” Proc. - 2nd World Congr. Comput. Commun. Technol. WCCCT
2017, pp. 223–226, 2017.
[3] D. Morocho, A. Morales, J. Fierrez, and R. VeraRodriguez, “Towards human-assisted signature
recognition: Improving biometric systems through attribute-based recognition,” ISBA 2016 - IEEE Int.
Conf. Identity, Secur. Behav. Anal., 2016
[4] R. Sa-Ardship and K. Woraratpanya, “Offline Handwritten Signature Recognition Using Polar-Scale
Normalization,” 8th Int. Conf. Inf. Technol. Electr. Eng. (ICITEE), Yogyakarta, Indones., pp. 3–7, 2016.
[5] Vinayak Jadhav, Nikhil kadam, Paresh Keluskar, Ayyaj Khan, “Artificial Neural Network Based
Signature Verification,” IOSR Journal of Computer Engineering (IOSR-JCE) 2016.
[6] T. Marušić, Ž. Marušić, and Ž. Šeremet, “Identification of authors of documents based on offline
signature recognition,” MIPRO, no. May, pp. 25–29, 2015.
[7] S. A. Angadi, S. Gour, and G. Bhajantri, “Offline Signature Recognition System Using Radon
Transform,” 2014 Fifth Int. Conf. Signal Image Process., pp. 56–61, 2014.
[8] S. A. Angadi and S. Gour, “Euclidean Distance Based Offline Signature Recognition System Using
Global and Local Wavelet Features,” 2014 Fifth Int. Conf. Signal Image Process., pp. 87–91, 2014.
[9] G. S. Prakash and S. Sharma, “Computer Vision & Fuzzy Logic based Offline Signature Verification
and Forgery Detection,” IEEE Int. Conf. Comput. Intell. Comput. Res., 2014.
[10] M. S. Shirdhonkar and M. B. Kokare, “Document image retrieval using signature as query,” 2011 2nd
Int. Conf. Comput. Commun. Technol. ICCCT-2011, pp. 66–70, 2011.
[11] S. Hangai, S. Yamanaka, and T. Hammamoto, “ONLINE SIGNATURE VERIFICATION BASED ON
ALTITUDE AND DIRECTION OF PEN MOVEMENT,” Proc. 15th Int. Conf. Pattern Recognition. ICPR-
2000, vol. 3, no. l, pp. 479–482, 2000.
IJCRT2205942 International Journal of Creative Research Thoughts (IJCRT) [Link] i28