Signature Recognition for Banking System

Abstract

The use of signatures for personal identification and verification is very common. Verifying signatures is necessary for many documents, including legal transactions and bank checks. The authentication of several papers using signatures is an extremely challenging and time-consuming job. As a result, systems for biometric personal verification and identification that rely on measurable, distinctive physical traits (such as fingerprints, hand geometry, faces, iris scans, or DNA) or behavioural traits have experienced an accelerating expansion (gait, voice etc.). Since tokens, passwords, pins, and other commonly used identity verification techniques have a number of fatal flaws and can\'t meet security requirements, the research looks into a feature that is far more often used: signature verification. We discuss the methods for authenticating signatures. We classify and outline the many techniques put forth for verification of signatures.

Introduction

I. INTRODUCTION

In addition to a well-liked topic of study in this fields of pattern recognition and image processing, signature verification is crucial to many applications, including access control, security, and privacy. The process of certifying someone based on their handwritten signature is known as signature verification.

Systems for verifying signatures come in two varieties [1].

Online Signature Verification System, which uses a tablet or other electronic device to collect details such as pressure, speed, and direction.

Verifying Offline Signatures System where the signature is written offline as well as the image scan is read before the images of the signature is already saved to verify it.

There are two methods that may be used for offline signature verification [2]. One type of signature verification is writer dependent, in which models of real and fake signatures are built for each writer. After that, a writer's test signature sample is contrasted with its own training dataset. The drawback of this strategy is that each new writer must have a model created in order to be confirmed. Forensic professionals employ the second method, known as writer-independent signature verification [2]. Since it doesn’t necessary to develop a model for each writer in order to validate their signatures, this method is seen to be the most practical. In this instance, a broad model is created using a few writers selected at random. However, because to the significant morphological variation across writers, writer-independent signature verification is a more challenging issue.

Strong research has been done on signature verification, and it is currently being investigated, especially in the offline form [3,4]. Due to the absence of a lot of dynamic information in the offline mode and online signature has demonstrated much greater validation rates from offline verification [5].

Online validation is therefore frequently more efficient. Most of the time, writing styles rather than a listing of alphabets, characters, and words are indicated by signatures [6]. The signature of a person frequently varies based on stress, emotion, time, exhaustion, etc. Even in signatures, variations are seen based on behaviours, geography, and physical and mental health [7]. The structures of this essay are as follows: Section I several forgeries types, while Section II explains the process used for offline signature verification.

Performance assessment metrics are introduced in Section III. Different signature verification methods are introduced in Section IV. The article is concluded in Section V.

A.  Several Forgeries Types

Any signature verification system's primary task is to determine if a signature is real or a fake. Three categories of forgery are used in signature verification systems: random, basic, and skilled [8]. When a signer does not include any information regarding their identity or signature style, it is called as a random forgery. Expert forgery is those where the fraudster is both fully aware of the signatories and the style of the original signature, as opposed to simple forgery in which the fraudster just knows the name of the signatory as well as the style of a signatures. The hardest to identify forgeries are those made by experts.

II. METHODOLOGY

 Four processes make up the majority of offline signature verification systems: Signature capture, pre-processing, feature extraction, enrolment, and verification. Figure 1 shows the system's overall layout.

A. Signature Capture

Scanners and cameras are used to gather data for signatures validation so that it is available in digital form.

B. Pre-processing

The obtained signature has to be normalised, scaled to the correct proportions, thinned, and background noise removed. This provides a template for a signature that can be used to extract features. The knowledge base stores the features that were extracted.

C. Feature Extraction

In order to deliver organised information in the form of observed data, the system collects certain characteristics during in the feature extraction phase. By removing qualities and traits from a given picture, it does this. Any quantitative amount might be a characteristic. However, as the test signature will ultimately be classified solely based on these features, the extracted characteristics will have a significant impact on how accurate the verification is. The two types that best characterise feature extraction techniques are globally feature extraction and locally feature extraction.

1. Globally features such as the width, height, and edge points of the signatures serve to represent the entire signature image. Since they are lower sensitive to signatures alterations and noise, these attributes would be adequate for random forgery but it would not give a high degree of an accuracy for expert forgery.
2. Because locally features define a considerably smaller portion of the signatures and extract more specific information from it, they are more reliable than globally features.

D. Enrolment and Verification

The knowledge base stores the extracted characteristics. The qualities of a person's signature vary depending on their emotional or mental state, among other variables. The decision thresholds needed for classification are determined by taking the training set's feature diversity into account. Application influences how the threshold is chosen. High thresholds are utilised for highly secure applications like the military and others, whereas moderate thresholds are used for other applications like banking. Test samples are compared with the templates in the database after a specific threshold is fixed to determine if a given is authentic or a fake.

III. PERFORMANCE ASSESSMENT MATRICES

The effectiveness of a signature verification system relies on how precisely it can distinguish between authentic and fake autographs [9]. Utilizing a variety of performance assessment metrics, such as the False Rejection Rate, False Acceptance Rate, and Equal Error Rate, various handwriting signature verification systems are evaluated for their effectiveness [10][11][12]. The FRR is the proportion of the total number of submitted valid test signatures to the amount of genuine testing signatures that the system rejected. The FAR measures the proportion of approved forgeries to all submitted forgeries. The FAR will usually rise whenever the decision threshold was modified to reduce the FRR, and vice versa. The calculated FAR and FRR graphs depend on the movable selected threshold. The illustrations that follow display common outcomes in both linear & logarithmic scale.

IV. RELATED WORK

Offline signature verification has seen the development of several strategies and tactics. The following discussion          examines a few practical techniques and ideal ways.

The broad aspects of the signature may be recovered using Sabourin's approach [13] at a low resolution, and the remaining features from the signature's distinctive regions can be extracted using a good resolution [6]. In the validation decision process, he employs locally and globally information as merely a feature vector. A technique that makes use of local granulometric sizes has been proposed by Sabourin [14]. A grid of rectangular retinas serves as the focal point of the chosen signature picture, which is activated by certain parts of the signature. Local shape descriptors are created via granulometric size distributions to reflect the quantity of signal activity activating each retina slightly beyond the attention grid. To identify random forgery, he has also utilised a closest neighbour & threshold-based classifier. Overall error rates were 0.03 percent as well as 1.1 percent, respectively, according to the data. Ismail [15] suggested a method for recognising off-line signatures where the border was represented in chain code defined by linked sequences that line segments with such a predetermined length and orientation. 2400 signature photographs were taken into account by the database. Seven distinct types of distances were measured using feature vectors derived from Eigen-signatures. The accuracy of the Manhattan measuring distance is 96,2%. Abbas [16] created a back - propagation algorithm network architecture to create an offline technique for recognising signatures. He used feed-forward neural networks and three different training methods. Vanilla, batch, and improved. FARs between 10 & 40% were noted for blatant frauds. A technique for off-line signatures was created by Edson [18]. He discovered the stroke lines within the signature image using the Hough transform. A defining characteristic of signatures would be the separation of the signature skeletal from the parameterized Higher dimensional space, which is accomplished via the Hough transform. That used a backward propagation neural network, the performance of the proposed technique is approximately calculated. The system's recognition rate is 95.24%. E.N. Zois developed a unique method for offline signatures detection and verification[25]. Segmentation of such a signature trace is achieved by using a window centred on the thicker image's centre of mass. The image is divided into sections to create a multidimensional feature representation that provides the handwritten image's useful spatial features. A hard margin svm is used for classification.

Conclusion

For the most recent methods used in offline signature validations in this work, we present a state-of-the-art. Although a variety of techniques are used in this area, accuracy needs to be highly increased, especially for complex forgery. Review and comparison of the FAR, FRR, as well as accuracy for numerous available approaches. Since the accuracy of the present technologies has not been very good, further research in offline signature verification is necessary. Future research may combine different classifiers to produce better validation results.

References

[1] Jain, F. Griess, and S. Connell, “On-line signature verification,” Pattern Recognition, vol. 35, no. 12, pp. 2963–2972, 2002. [2] D. Bertolinia, L. Oliveirab, E. Justiona, and R. Sabourin, “Reducing Forgeries in Writer Independent Off-line Signature Verification through ensemble of Classifiers”. Pattern Recognition, vol. 43, January 2010, pp. 387-396. [3] Hai Rong Lv, Wen Jun Yin, Jin Dong, “Off-line Signature Verification based on deformable grid partition and Hidden Markov Models”, ICME2009, pp. 374-377. [4] S. Chen and S. Srihari, “Use of Exterior Contour and Shape Features in Off-line Signature Verification”, ICDAR- 2005, pp. 1280-1284. [5] M. Kalera, S. Srihari, and A. Xu. “Offline signature verification and identification using distance statistics”, IJPRAI- 2004, pp.1339-1360. [6] B. Fang, C.H. Leung, Y.Y. Tang, K.W. Tse, P.C.K. Kwok and Y.K. Wong, \"Off-line signature verification by the tracking of feature and stroke positions\",Pattern Recognition, 2003, pp 91-101. [7] R. Abbas and V. Ciesielski, “A Prototype System for Off-line Signature Verification Using Multilayered Feed Forward Neural Networks”, February 1995. [8] E. J. R. Justino, F. Bortolozzi, R. Sabourin, “Off-line signature verification using HMM for random simple and skilled forgeries”, Proc. 6th Intl. Conf. On Document Analysis and Recognition, 2001, pp. 450- 453. [9] B. Majhi, Y. Reddy, D. Babu, “Novel Features for Off-line Signature Verification”, International Journal of Computers,Communications & Control Vol.I (2006), No. 1, pp. 17-24. [10] H. Baltzakis, N. Papamarkos, “A new signature verification technique based on a two-stage neural network classifier”, Engineering Applications of Artificial Intelligence 14 (2001) [11] M.K kalera, S. Shrihari, “Offline Signature Verification and Identification Using Distance Statistics”, International Journal of Pattern Recognition and Artificial Intelligence Vol. 18, No. 7 (2004) 1339-1360, World Scientific Publishing Company [12] 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. [13] R. Sabourin, G. Genest, “An extended-shadow-code-based approach for off-line signature verification: Part I. Evaluation of the bar mask definition”, Proc. Of 12th ICPR, Jerusalem, Israel, 1994, pp. 450-453. [14] R. Sabourin, G. Genest, F. J. Preteux, “Off-Line Signature Verification by Local Granulometric Size Distributions”, IEEE Trans. Pattern Anal. Mach. Intell. 19 (9) (1997), pp. 976-988. [15] I.A. Ismail, M.A.Ramadan, T. S. El-Danaf and A.H. Samak, “An Efficient Off line Signature Identification Method Based On Fourier Descriptor and Chain Codes”, 2010, (IJCSNS-2010),VOL.10 No.5, pp.29-35. [16] R. Abbas, “Back propagation Neural Network Prototype for off line signature verification”, thesis Submitted to RMIT,2003. [17] Edson J. R. Justino, Abdenaimel Yacoubi, Flaviob Ortolozzi, Roberts Abourin, “An Off-Line Signature Verification System Using Hidden Markov Model and Cross-Validation”, IEEE Int. Workshop on Neural Networks for Signal Processing,pp. 859–868, 2000. [18] E. N. Zois, A.Nassiopoulos, K. Tselios, E. Siores, G. Economou “OffLine Signature Verification Using Two Step Transitional Features”, MVA2011 IAPR Conference on Machine Vision Applications, June 13- 15, 2011, Nara, JAPAN pp 295-298.

Copyright

Copyright © 2022 Madhu K N, Mrs. Bhavana G. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.