2023 Third International Conference on Secure Cyber Computing and Communication (ICSCCC)

Estimating the Effectiveness of CNN and Fuzzy

Logic for Signature Verification
2023 Third International Conference on Secure Cyber Computing and Communication (ICSCCC) | 979-8-3503-0071-0/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICSCCC58608.2023.10176476

Ishu Priya Dr. Nisha Chaurasia

Department of Computer Science and Engineering Department of Information Technology
Dr. B R Ambedkar National Institute of Technology Dr. B R Ambedkar National Institute of Technology
Jalandhar, India Jalandhar, India
[Link].21@[Link] chaurasian@[Link]

Abstract—Each individual has a distinctive signature which is

primarily used for personal authentication as well as the
confirmation of significant papers or formal processes. A Sig-
nature, a form of bio-metric, is a method of authenticating the
identity of an individual, thus, helping manage security. This
reduces the possibility of forgery during the authentication
process, and aids in avoiding any discrepancies during the signing
process. But for a huge number of files, offline signature-based
detection is ineffective and sluggish. Herein comes the need for
automated signature verification systems. These systems aim at
verifying whether a given signature image is real or fake. This Fig. 1. Sample Images for Offline Signatures [36]
paper provides a comprehensive survey of literature on various
techniques used for handwriting verification including signatures.
The respective findings of various methods have been discussed,
highlighting the latest trends of research and their potential Fig. 1 gives a glimpse of what hand-written signatures look
applications. The paper aims to present state-of-the-art methods like. Signatures done in Bengali and Hindi languages are
for future development thus, serving as a helpful means for depicted in this figure.
researchers carrying out study in signature verification and
related topics. A. Need for an Automated Signature Verifier
Index Terms—Machine Learning, Deep Learning, Neural Net- On many different kinds of documents, like legal doc-
work, Convolutional Neural Network, Fuzzy Logic uments, bank cheques, etc., handwritten signatures can be
I. INTRODUCTION spotted. Given the large number of these documents, which is
still growing day-by-day, an automated system for signature
Along with being a well-liked topic of study in the fields of verification is preferred which need to be highly dependable
image and pattern recognition, signature verification is also and accurate. Jungpil Shin and Tomomi Kikuchi [10] worked
crucial to numerous applications which cover security and on proposing an algorithm to build a signature verifier using
privacy concerns. In recent times, offline signature forgery de- fuzzy set theory.
tection has drawn a lot of attention from researchers. In under-
lying facts, the system proposed for detecting forgery should B. Role of a Deep Neural Network
be able to discern signatures from signed papers and verify Since, it is a well-known fact that ANNs do not work
them. But there are numerous difficulties on the horizon such efficiently on image data as it requires the 2D images to be
as low-quality signature images which are often difficult to converted into 1D space, which is a difficult task for ANN
find. Apart from this, it may be a difficult task to verify to perform hence, Convolutional Neural Networks (CNNs) are
a cropped signature as it could be containing immaterial the best suitable method for image classification problems[24].
background context from the document. Besides some fake Thus, the proposed system makes use of a CNN classifier
signatures closely resemble the real ones, making verification which is discussed in the upcoming sections.
a harder problem to deal with.
There are two types of signatures based on their acqui- C. Role of a Fuzzy Logic System
sition namely online and offline. Offline signatures are 2D In general, fuzzy logic is nothing more than a set of rules.
digital images, usually encapsulated using document scanning Fuzzy sets comprise vague or imprecise data [28]. A fuzzy
whereas online signatures are acquired using certain special logic system tells the degree of truth instead of the actual
accession devices. This research mainly focuses on offline truth which in the simplest terms means that a fuzzy logic
signatures. Despite being one of the most fundamental and defines that a statement may or may not be true. Because of
widely used methods of authentication, yet determining the these perks of fuzzy logic, many researchers have worked on
verity or authenticity of a signature is exceptionally difficult. studying the application of fuzzy logic systems in the field of
hand-written image recognition [1].
979-8-3503-0071-0/23/$31.00 ©2023 IEEE 367
Authorized licensed use limited to: Sungkyunkwan University. Downloaded on August 28,2023 at 02:57:44 UTC from IEEE Xplore. Restrictions apply.
 2023 Third International Conference on Secure Cyber Computing and Communication (ICSCCC)

In brief, to comment on why there is a need to explore fuzzy fuzzy sets thus resulting in improved efficiency and accuracy.
logic when we already have the concept of crisp (boolean) Similarly, M Hanmandlu et al [3] worked on formulating a
logic is because fuzzy logic doesn’t require explicit inputs research method recognized as the box method in the field
to work on as these systems are capable of incorporating of feature extraction so as to recognize handwritten charac-
imprecise data [11]. One of the main advantages of using ters. The dataset chosen for conducting the experiment was
fuzzy logic in signature verification is its ability to extract CEDAR CDROMI. The feature recognition was performed
features that are robust to variations in the signature such as using back-propagation (BPNN) and fuzzy logic, yielding a
size, rotation, and slant. 100% recognition rate.
Fig. 2 depicts the basic difference between crisp logic and Y. Alginahi et al [5] used a novel set of fuzzy-descriptive
fuzzy logic. features to process optical code, implemented using the cross-
correlation for character classification. The study was con-
ducted using the MICR codes located at the bottom of bank
cheques. The results drawn showed that it was feasible to
attain 100% recall for the training patterns and apart from
this, 100% recognition rate for warped and skewed patterns
was also achieved.
Taking into consideration the power that fuzzy logic of-
fers, M. Soleymani Baghshah et al [6] proposed a novel
fuzzy-based methodology to recognize Persian hand-written
characters. Fuzzy linguistic modeling approach was used to
accomplish the character parameters representation and these
features were used to rule out the token shapes. The model
Fig. 2. Fuzzy Logic v/s Boolean Logic
yielded a 95% recognition rate. Elaheh Dehghani and Mohsen
Ebrahimi Moghaddam [9] worked on Persian Signature dataset
Although the NN is a well-capable classifier, the vitality of and SUBCORPUS-100-MCYT to mount a verification system
infusing fuzzy logic is that a fuzzy logic system is heavily for online signatures using global information and ANFIS
dependent on the generation of rules and membership func- model. Mohammed Zeki Khedher and Dr. Ghayda Al-Talib [8]
tions to yield output. But to generate these rules, domain also worked on handwriting recognition for Arabic sub-words
expertise is required, which might not be available always. using fuzzy logic. Membership functions were generated to
Hence, to deal with this limitation of the fuzzy classifier, a recognize these patterns using feature extraction.
neural network (NN) can be trained which can then be used G.E.M.D.C. Bandara et al [4] came forward to develop
to generate these rules [26] [27]. Several researchers conducted a portable, efficient, and relatively short algorithm so as
detailed surveys of work done so far in this domain by making to replicate human intellect targetting hand-written alpha-
use of NNs and fuzzy logic-based systems, thus, helping them numeric characters. It took into account fuzzy feature extrac-
get better insights into their research [16] [21] [25]. These tion methods to examine the uncertainties and vagueness of
surveys also helped the researchers to strengthen their much- handwritten characters and concluded that the given method
valuable contributions by helping others to readily understand outperformed the traditional methodologies by yielding better
the ongoing trends in this domain. results. Similarly, K.B.M.R. Batuwita and G.E.M.D.C. Ban-
The rest of the paper is organized in the given manner: dara [7] together worked on developing a fuzzy-based nu-
• Section II comprises the related work done till date in meric character recognition system. The authors binarized and
this domain. skeletonized the input character images and then performed
• Section III describes the proposed methodology and the isolation and segmentation on the characters individually.
results inferred so far in this research. In addition to fuzzy logic, there are several other techniques
• Section IV discusses the main motivation for using a used in the field of hand-written image classification which are
CNN model with a fuzzy classifier. discussed in the upcoming sub-sections.
• At last, Section V concludes the entire paper.

II. REVIEW OF LITERATURE B. Neural Networks and Computer Vision

A. Fuzzy Logic
Image classification is a common application of NNs and
Yan Solihin et al [2] proposed an image enhancement Image processing. Over the years, there has been significant
method to extract handwritten images from a document by amount of research done in this domain and many researchers
making use of fuzzy logic. This method was put forward to have applied ML, DL, image processing etc. for hand-written
refine the pre-processing of handwritten documents. It suggests image classification. Some of these techniques used have been
fuzzy-based if-then rules which are mainly put-up onto two described using the table 1.

368
Authorized licensed use limited to: Sungkyunkwan University. Downloaded on August 28,2023 at 02:57:44 UTC from IEEE Xplore. Restrictions apply.
 2023 Third International Conference on Secure Cyber Computing and Communication (ICSCCC)

TABLE I
COMPARISON OF VARIOUS NNS AND IMAGE PROCESSING TECHNIQUES

Year Paper Title Author Methodology Findings

2007 Signature envelope curvature Tai-Ping Zhang et al Feature extraction done using rotation- The AER (average error rate) came
descriptor for offline signature [12] invariant curvature sequences of signa- out to be 17.17%.
verification ture envelope.
2012 A Survey for Handwritten Sig- Ahmad Sanmorino, Presented survey on several techniques Concluded that NNs and HMM
nature Verification Setiadi Yazid [13] such as NNs, HMM, Fuzzy Logic, pre- were better options from imple-
processing techniques for feature ex- mentation perspective.
traction etc.
2013 Off-line Signature Verification K. V. Lakshmi, Seema Used multi-layered NN, pre-processing Fault Acceptance Rate: 12% and
Using Neural Networks Nayak [14] was done using “im toolkit”. Fault Rejection Rate: 8%.

2013 A novel signature verification Md. Iqbal Quraishi et ANN for classification; Log Polar and The model accuracy came out to be
and authentication system us- al [15] Ripplet-II Transformation method and around 96.15%.
ing image transformations and Fractal Dimension for descriptor ex-
Artificial Neural Network traction.
2013 ICDAR2013 Competitions Md. Imran Malik et al Presented the signature modality out- Signature verifiers gave
on Signature Verification and [16] puts in terms of likelihood ratio and satisfactory performance on
Writer Identification for On- reported ICDAR2013 competitions re- various datasets whereas writer
and Offline Skilled Forgeries sults. identifiers couldn’t perform well.
(SigWiComp2013)
2013 Off-Line Skilled Forgery De- Nivedita Yadav et al SFS was combined with k-NN to se- The model yielded an accuracy of
tection on Handwritten De- [17] lect distinct primitives, HOG was used 93.33%.
vanagri Script for object detection, and RBF-SVM for
recognition purpose.
2014 Computer Vision & Fuzzy Gautam. S. Prakash, Artificial Neural Network, Fuzzy The model gave satisfactory results
Logic based Offline Signature Shanu Sharma [11] Logic, MATLAB but still requires certain improve-
Verification and Forgery De- memts.
tection
2014 GMM For Offline Signature Neerja Arora et al [18] GMM was used for taining the model The model accuracy came out to be
Forgery Detection and Euclidean Distance was calculated 87.14%.
to define the acceptance range between
the signatures.
2016 “A Statistical Approach for Of- A.B.M. Ashikur Rah- Applied Harris Corner Detector, fea- The proposed system outperformed
fline Signature Verification us- man et al [19] ture descriptor was calculated using the the other methodologies used.
ing Local Gradient Features SIFT operator.
2016 “On-line signature verification N. Aqili et al [20] Used point pattern matching, and DTW Rejecting Tendency (real sign):
using Point Pattern Matching algorithm was applied to draw corre- 10% and Accepting Tendency (fake
Algorithm spondence between 2 signals. sign): 30%.

2018 Handwritten Signature Anastasia Beresneva et Put-forth the work done so far on hand- The experiments conducted yielded
Attributes for its Verification al [21] written signatures WT, Discrete Radon, that DWT was the most effective
FT and other methods. technique in this case.

2018 Handwritten Text Recognition Rajib Ghosh et al [22] Used methods like cropping, segmen- The method accuracy was 89.73%.
in Bank Cheques tation, feature extraction using GLCM
and HOG methods, and isolated char-
acter recognition using SVM classifier.
2018 Symbolic Representation Snehal K. Jadhav, M. Local Binary Pattern method was used The model accuracy was 90.36%.
Model for Off-line Signature K. Chavan [23] for feature extraction, and k-NN for
Verification classification.
2020 Handwritten Digit Recognition Chao Zhang et al [24] Proposed a CNN model for handwrit- Accuracy was 97.3%.
Based on Convolutional Neu- ing pattern recognition, regression al-
ral Network gorithm used was Softmax.

C. Neural Networks and Fuzzy Logic both techniques can be applied in combination to create a
robust signature verification system that can detect and prevent
In terms of verifying signatures, both fuzzy logic and neural forgery.
networks offer advantages and disadvantages. NNs can be Thus, many researchers, considering the merits and de-
very efficient at identifying various patterns in complex data, merits of both these techniques, worked on proposing such
however they require a large (significant) amount of training systems that inherit the features of both a NN and fuzzy-logic.
data and may not always be able to explain how they actually
arrived at a decision. On the other hand, fuzzy logic can deal Some of such methodologies that have been worked upon
with uncertainty and ambiguities but may need more fine- over the years by several researchers have been summarized
tuning and domain-specific expertise to be effective. Overall, in table 2.

369
Authorized licensed use limited to: Sungkyunkwan University. Downloaded on August 28,2023 at 02:57:44 UTC from IEEE Xplore. Restrictions apply.
 2023 Third International Conference on Secure Cyber Computing and Communication (ICSCCC)

TABLE II
COMPARISON OF NEURO-FUZZY TECHNIQUES

Year Paper Title Author Methodology Findings

2022 Handwritten Signature Veri- Sudharshan Duth P et Comparison of VGG16, VGG19 and VGG19 outperformed the other
fication System using Deep al [34] ResNet50 on SigComp2009 data and two models.
Learning some raw input
2022 The Dual-Fuzzy Convolutional Wei Zhou et al [35] Hybrid DualFuzzy CNN, Probabilistic Optimized algorithm, helps relax
Neural Network to Deal With hestitant fuzzy elements the requirements of input, gives
Handwritten Image Recogni- better performances
tion
2021 A Robust Model for Hand- Ayush Kumar Agrawal Image classification, CNN, Keras, RM- Accuracy on MNIST dataset:
written Digit Recognition us- et al [33] Spropoptimizer around 99.06%.
ing Machine and Deep Learn-
ing Technique
2020 Offline Signature Recognition Jivesh Poddar et al CNN, Crest-Trough method, Harris cor- The model gave an accuracy of
and Forgery Detection using [30] ner detection and SURF for forgery about 85-89%.
Deep Learning classification
2020 Handwritten Form Recogni- Narayana Darapaneni Sequence-to-Sequence implementation, The system gave better accuracy
tion Using Artificial Neural et al [32] LSTM on IAM dataset, CNN, contour than using traditional CNN without
Network detection contours.

2020 Adaptive Fuzzy Network Rishil Shah [31] Adaptive fuzzy network-based CNN for The given method outperformed
based Transfer Learning for classification and image segmentation, the existing methods of transfer
Image Classification Transfer Learning learning.

2019 Interpretable Deep Convolu- Mojtaba Yeganejou et Rocchio’s algorithm, CNN, fuzzy clas- Performed better and introduced an
tional Fuzzy Classifier al [29] sifier effective mechanism of interpreta-
tion.

2019 A Deep neuro-fuzzy network O. Yazdanbakhsh et al Deep Neuro-Fuzzy network, TSK fuzzy Contributed that deep structures
for image classification [28] model, SIMO rule based system image when combined with fuzzy systems
classification. can be used in image analysis.

2019 Fuzzy based Pooling in Con- Teena Sharma et al Fuzzy logic using CNN having a single Introduced a novel pooling method
volutional Neural Network for [27] layer, Gaussian Membership function, to facilitate dimension reduction;
Image Classification Type-1 fuzzy set better accuracy.

2018 A Convolutional Fuzzy Neural Kseniya P. Korshunova CNN, fuzzy based logic system, VGG The hybrid model gave better ac-
Network for Image Classifica- [26] model for classification curacy than traditional CNN.
tion
2017 A Survey of Offline Handwrit- Ajay Indian et al[25] Reviewed the research work done by The survey reported that the most
ten Hindi Character Recogni- various other researchers on Hand- widely used character pattern is
tion written character recognition. that of Hindi language.

A pie chart has been depicted below in Fig. 3 to provide a III. PROPOSED METHODOLOGY
rough idea of the statistics of the various algorithms that have
been reviewed in the literature survey done for carrying out A Fuzzy-based CNN model is the amalgamation of the CNN
this research. and fuzzy logic. The methodology that our research proposes
From the comparative analysis done in Fig. 3, it can be is carried out in the following manner:
drawn out that there has been some amount of work done • Preprocessing: The signature images in the dataset BH-
using the hybrid model systems but still there is a large scope Sig260 are pre-processed to remove any noise or un-
for more research that can be carried out in this field. wanted artifacts. This step includes image enhancement,
normalization, and segmentation.
• ROI Extraction:
– Region of Interest (ROI) is extracted by applying
FCM clustering to the pre-processed signature im-
age, which contains the signature. The pixels are
grouped by the clustering algorithm on the basis of
their intensity values and each pixel is assigned a
degree of membership to different clusters.
– Based on the membership degree, the pixels in the
ROI are selected to the cluster that corresponds to
Fig. 3. Comparative analysis of methods used in this paper the signature pixels. The ROI is then extracted from

370
Authorized licensed use limited to: Sungkyunkwan University. Downloaded on August 28,2023 at 02:57:44 UTC from IEEE Xplore. Restrictions apply.
 2023 Third International Conference on Secure Cyber Computing and Communication (ICSCCC)

the signature image. decision-making processes. This makes it especially helpful in

– The pixel-wise membership on input image is cal- real-world situations where there is inherent uncertainty and
culated by the FCM clustering algorithm, which in clear rules are hard to articulate [2] [4] [5]. On the other hand,
turn yields a segmented image. CNNs are a kind of Deep Learning model that have the ability
• Feature Extraction: To extract discriminative features to automatically learn hierarchical representations of visual
of the signature, the ROI is fed into a pre-trained CNN data for classification and other related tasks.
model. The CNN learns a set of features that capture Though there are several techniques that can be combined with
the key characteristics of the signature, such as shape, fuzzy logic to develop an effective model but CNN, being the
texture, and curvature. Finally, the CNN model classifies state-of-the-art technique in field of image classification, offers
whether the given signature image is real or fake. an edge over the other methodologies. These two techniques
Fig. 6 depicts our proposed methodology for designing the when combined together complement each other’s benefits
hybrid model. [29] [31] [35].

V. CONCLUSION

This paper presents a detailed literature survey on various

methodologies used in the field of signature verification.
From the literature survey done so far, it could be inferred
that CNNs are the most widely used neural networks in the
domain of image classification. But these models sometimes
fail to recognize uncertainties in the images. Hence, a fuzzy
classifier when merged with a neural network tends to
perform better than the conventional methodologies as fuzzy
logic systems are highly capable of handling uncertainties
in the data. To strengthen the statement, we implemented
our proposed methodology and the initial results are quite
promising, with an accuracy of 86.75%, giving a good start
in this research, leaving a scope of further improvement and
research in this field.

REFERENCES
Fig. 4. Architecture of Fuzzy Logic-based CNN model
[1] P. D. Gader and J. M. Keller, ”Fuzzy methods in handwriting recog-
We evaluated the performance of our model based on several nition: an overview,” Proceedings of North American Fuzzy Infor-
mation Processing, Berkeley, CA, USA, 1996, pp. 137-141, doi:
parameters which have been described in table 3. 10.1109/NAFIPS.1996.534718.
[2] Y. Solihin, C. G. Leedham and V. K. Sagar, ”A fuzzy based handwriting
TABLE III extraction technique for handwritten document preprocessing,” Proceed-
RESULTS OBTAINED ings of Digital Processing Applications (TENCON ’96), Perth, WA, Aus-
tralia, 1996, pp. 927-932 vol.2, doi: 10.1109/TENCON.1996.608472.
Parameter Result [3] M. Hanmandlu, K.R.M. Mohan, S. Chakraborty, “Fuzzy logic based
Accuracy 86.75% handwritten character recognition”, Proceedings 2001 International Con-
Precision 92.5% ference on Image Processing ([Link].01CH37205), October 2001
Recall 76.8% [4] G. E. M. D. C. Bandara, S. D. Pathirana and R. M. Ranawana, ”Use
F1-score 83.9% of fuzzy feature descriptions to recognize handwritten alphanumeric
characters,” 2002 IEEE World Congress on Computational Intelligence.
2002 IEEE International Conference on Fuzzy Systems. FUZZ-IEEE’02.
Proceedings (Cat. No.02CH37291), Honolulu, HI, USA, 2002, pp. 1586-
1591 vol.2, doi: 10.1109/FUZZ.2002.1006743.
IV. SIGNIFICANCE OF CNN AND FUZZY LOGIC
[5] Y. Alginahi, I. El-Feghi, M. Ahmadi and M. A. Sid-Ahmed, ”Optical
character recognition system based on a novel fuzzy descriptive fea-
CNN is one of the most widely used techniques for image tures,” Proceedings 7th International Conference on Signal Processing,
classification. It works on convolutional layers that are used 2004. Proceedings. ICSP ’04. 2004., Beijing, China, 2004, pp. 926-929
for feature extraction from the image. These features are then vol.2, doi: 10.1109/ICOSP.2004.1441471.
[6] M. Soleymani Baghshah, S. Bagheri Shouraki and S. Kasaei, “A Novel
combined with several other layers that are used to represent Fuzzy Approach to Recognition of Online Persian Handwriting”, 5th
the image in more abstract manner, thus, making it easier International Conference on Intelligent Systems Design and Applications
for the model to recognize much more complex patterns [29] (ISDA’05), September 2005
[31] [32]. Fuzzy Logic is a mathematical model that deals [7] K. B. M. R. Batuwita and G. E. M. D. C. Bandara, ”Fuzzy Recognition
of Offline Handwritten Numeric Characters,” 2006 IEEE Conference on
with ambiguity and imprecision in a manner reminiscent of Cybernetics and Intelligent Systems, Bangkok, Thailand, 2006, pp. 1-5,
human reasoning. It can flexibly simulate intricate linkages and doi: 10.1109/ICCIS.2006.252356.

371
Authorized licensed use limited to: Sungkyunkwan University. Downloaded on August 28,2023 at 02:57:44 UTC from IEEE Xplore. Restrictions apply.
 2023 Third International Conference on Secure Cyber Computing and Communication (ICSCCC)

[8] M. Z. Khedher and G. Al-Talib, ”A fuzzy expert system for recognition [24] C. Zhang, Z. Zhou and L. Lin, ”Handwritten Digit Recognition
of handwritten arabic sub-words,” 2007 9th International Symposium on Based on Convolutional Neural Network,” 2020 Chinese Automa-
Signal Processing and Its Applications, Sharjah, United Arab Emirates, tion Congress (CAC), Shanghai, China, 2020, pp. 7384-7388, doi:
2007, pp. 1-4, doi: 10.1109/ISSPA.2007.4555355. 10.1109/CAC51589.2020.9326781.
[9] E. Dehghani and M. E. Moghaddam, ”On-line signature verification [25] A. Indian and K. Bhatia, ”A survey of offline handwritten Hindi char-
using ANFIS,” 2009 Proceedings of 6th International Symposium on acter recognition,” 2017 3rd International Conference on Advances in
Image and Signal Processing and Analysis, Salzburg, Austria, 2009, pp. Computing, Communication & Automation (ICACCA) (Fall), Dehradun,
546-549, doi: 10.1109/ISPA.2009.5297687. India, 2017, pp. 1-6, doi: 10.1109/ICACCAF.2017.8344697.
[10] J. Shin and T. Kikuchi, ”On-Line Signature Evaluation Using Fuzzy Set [26] K. P. Korshunova, ”A Convolutional Fuzzy Neural Network for Image
Theory,” 2013 27th International Conference on Advanced Information Classification,” 2018 3rd Russian-Pacific Conference on Computer Tech-
Networking and Applications Workshops, Barcelona, Spain, 2013, pp. nology and Applications (RPC), Vladivostok, Russia, 2018, pp. 1-4, doi:
273-277, doi: 10.1109/WAINA.2013.21. 10.1109/RPC.2018.8482211.
[11] G. S. Prakash and S. Sharma, ”Computer vision & fuzzy logic based [27] T. Sharma, V. Singh, S. Sudhakaran and N. K. Verma, ”Fuzzy
offline signature verification and forgery detection,” 2014 IEEE In- based Pooling in Convolutional Neural Network for Image Classifica-
ternational Conference on Computational Intelligence and Comput- tion,” 2019 IEEE International Conference on Fuzzy Systems (FUZZ-
ing Research, Coimbatore, India, 2014, pp. 1-6, doi: 10.1109/IC- IEEE), New Orleans, LA, USA, 2019, pp. 1-6, doi: 10.1109/FUZZ-
CIC.2014.7238363. IEEE.2019.8859010.
[12] T. P. Zhang, B. Fang, B. Xu, H. X. Chen, M. Chen and Y. [28] O. Yazdanbakhsh, S. Dick, “A Deep Neuro-Fuzzy Network for Image
Y. Tang, ”Signature envelope curvature descriptor for offline signa- Classification”, Neural and Evolutionary Computing, December 2019,
ture verification,” 2007 International Conference on Wavelet Anal- [Link]
ysis and Pattern Recognition, Beijing, 2007, pp. 1262-1266, doi: [29] M. Yeganejou, S. Dick and J. Miller, ”Interpretable Deep Convolutional
10.1109/ICWAPR.2007.4421628. Fuzzy Classifier,” in IEEE Transactions on Fuzzy Systems, vol. 28, no.
7, pp. 1407-1419, July 2020, doi: 10.1109/TFUZZ.2019.2946520.
[13] A. Sanmorino and S. Yazid, ”A survey for handwritten signature veri-
fication,” 2012 2nd International Conference on Uncertainty Reasoning [30] J. Poddar, V. Parikh, S. K. Bharti, ”Offline Signature Recogni-
and Knowledge Engineering, Jalarta, Indonesia, 2012, pp. 54-57, doi: tion and Forgery Detection using Deep Learning,” Procedia Com-
puter Science, vol. 170, 2020, Pages 610-617, ISSN 1877-0509,
10.1109/URKE.2012.6319582.
[Link]
[14] K. V. Lakshmi and S. Nayak, ”Off-line signature verification using [31] R. Shah, ”Adaptive Fuzzy Network based Transfer Learning for Im-
Neural Networks,” 2013 3rd IEEE International Advance Comput-
age Classification,” 2020 IEEE International Students’ Conference on
ing Conference (IACC), Ghaziabad, India, 2013, pp. 1065-1069, doi:
Electrical,Electronics and Computer Science (SCEECS), Bhopal, India,
10.1109/IAdCC.2013.6514374.
2020, pp. 1-4, doi: 10.1109/SCEECS48394.2020.155.
[15] M. I. Quraishi, A. Das and S. Roy, ”A novel signature verification [32] N. Darapaneni et al., ”Handwritten Form Recognition Using Artificial
and authentication system using image transformations and Artificial Neural Network,” 2020 IEEE 15th International Conference on Industrial
Neural Network,” 2013 World Congress on Computer and Information and Information Systems (ICIIS), RUPNAGAR, India, 2020, pp. 420-
Technology (WCCIT), Sousse, Tunisia, 2013, pp. 1-6, doi: 10.1109/WC- 424, doi: 10.1109/ICIIS51140.2020.9342638.
CIT.2013.6618680. [33] A. K. Agrawal, A. K. Shrivas and V. k. Awasthi, ”A Robust Model
[16] M. I. Malik, M. Liwicki, L. Alewijnse, W. Ohyama, M. Blumenstein and for Handwritten Digit Recognition using Machine and Deep Learning
B. Found, ”ICDAR 2013 Competitions on Signature Verification and Technique,” 2021 2nd International Conference for Emerging Tech-
Writer Identification for On- and Offline Skilled Forgeries (SigWiComp nology (INCET), Belagavi, India, 2021, pp. 1-4, doi: 10.1109/IN-
2013),” 2013 12th International Conference on Document Analysis CET51464.2021.9456118.
and Recognition, Washington, DC, USA, 2013, pp. 1477-1483, doi: [34] D. P. Sudharshan and R. N. Vismaya, ”Handwritten Signature Verifica-
10.1109/ICDAR.2013.220. tion System using Deep Learning,” 2022 IEEE International Conference
[17] N. Yadav, S. Chaudhury and P. Kalra, ”Off-line skilled forgery de- on Data Science and Information System (ICDSIS), Hassan, India, 2022,
tection on handwritten Devanagri script,” 2013 Fourth National Con- pp. 1-5, doi: 10.1109/ICDSIS55133.2022.9915833.
ference on Computer Vision, Pattern Recognition, Image Process- [35] W. Zhou, M. Liu and Z. Xu, ”The Dual-Fuzzy Convolutional Neural
ing and Graphics (NCVPRIPG), Jodhpur, India, 2013, pp. 1-4, doi: Network to Deal With Handwritten Image Recognition,” in IEEE
10.1109/NCVPRIPG.2013.6776172. Transactions on Fuzzy Systems, vol. 30, no. 12, pp. 5225-5236, Dec.
[18] N. Arora, A. Kumar and C. Jain, ”GMM for offline signature forgery 2022, doi: 10.1109/TFUZZ.2022.3170657.
detection,” 2014 5th International Conference - Confluence The Next [36] [Link]
Generation Information Technology Summit (Confluence), Noida, India, datasets
2014, pp. 576-581, doi: 10.1109/CONFLUENCE.2014.6949044.
[19] A. B. M. A. Rahman, G. Mostaeen and M. H. Kabir, ”A statistical
approach for offline signature verification using local gradient features,”
2016 2nd International Conference on Electrical, Computer & Telecom-
munication Engineering (ICECTE), Rajshahi, Bangladesh, 2016, pp. 1-4,
doi: 10.1109/ICECTE.2016.7879570.
[20] N. Aqili, A. Maazouzi, M. Raji, A. Jilbab, S. Chaouki and A. Ham-
mouch, ”On-line signature verification using point pattern matching
algorithm,” 2016 International Conference on Electrical and Information
Technologies (ICEIT), Tangiers, Morocco, 2016, pp. 410-413, doi:
10.1109/EITech.2016.7519631.
[21] A. Beresneva, A. Epishkina and D. Shingalova, ”Handwritten signa-
ture attributes for its verification,” 2018 IEEE Conference of Russian
Young Researchers in Electrical and Electronic Engineering (EICon-
Rus), Moscow and St. Petersburg, Russia, 2018, pp. 1477-1480, doi:
10.1109/EIConRus.2018.8317376.
[22] R. Ghosh, C. Panda and P. Kumar, ”Handwritten Text Recognition in
Bank Cheques,” 2018 Conference on Information and Communication
Technology (CICT), Jabalpur, India, 2018, pp.1-6, doi: 10.1109/INFO-
COMTECH.2018.8722420.
[23] S. K. Jadhav and M. K. Chavan, ”Symbolic Representation Model for
Off-Line Signature Verification,” 2018 9th International Conference on
Computing, Communication and Networking Technologies (ICCCNT),
Bengaluru, India, 2018, pp. 1-5, doi: 10.1109/ICCCNT.2018.8494145.

372
Authorized licensed use limited to: Sungkyunkwan University. Downloaded on August 28,2023 at 02:57:44 UTC from IEEE Xplore. Restrictions apply.