Design and implementation of deep learning strategy based smart signature verification System

https://doi.org/10.1016/j.micpro.2020.103119Get rights and content

Abstract

The signature verification is broadly used for personal identification. The person is identified automatically using signature verification method to avoid forgery persons. The signature verification is classified into the static method and the dynamic method. The static verification method is based on stored images and the dynamic verification method is based on dynamic features of the signature. The integer wavelet transformation method is used to identify the breath and height ratio of the signature features. In addition to that spurious noise also removed before extracting the signature feature. And the signature is isolated from the background of the images. The extracted feature is analyzed using integer wavelet transformation and a neural network is selected to decide according to that original and forgery signature. As compared with the conventional system the proposed found to be about 20% error ratio. The database SVC2004 is selected to verify the signature.

Introduction

Biometric signature verification system plays a vital role in human identification system. Generally human characteristics are classified into movable types and immovable types. Almost the movable systems are called static systems and immovable systems are called dynamic systems [1].Human physical characteristic like face identification, iris verification, fingerprint authentication are related to static system. The behavioral characteristics like signing their signature, pen holding, position of writing, interval taken to sign the signature are considered as a dynamic systems [2]. So the dynamic system is given more accurate result than the static system. And dynamic system replaces password and fingerprint authentication. So the dynamic system is globally accepted for biometric verification and authentication method [3].

Similarly the feature of the signature is classified into global feature selection method and local feature selection method [4]. In global feature selection method the entire individual parameters velocity, azimuth angle, pressure and acceleration of the signature are calculated in terms of average value or max value [5]. But in local feature selection method only single points are calculated to authenticate the output.

By using the proposed back propagation based neural network is selected to decide according to that original and forgery signature. From the database SVC2004 is trained during the process and it will be classified under the test result condition

The noises are present in the signature are removed after the signature send to preprocessing steps. The features are extracted in preprocessed image and different signatures are analyzed. Section 2 deals with the related works. Section 3 clarifies the preprocessing steps and the features that are extracted followed by the verification procedure in Section 4. Implementation details and simulation results are listed in Section 5. The conclusion follows in Section 6.

Section snippets

Related work

SVC2004 and SUSIG databases has been selected for signature verification. In each database signature many parameters are x coordinate, y coordinate, azimuth angle, altitude and pressure of the data is extracted and calculated using derivative of the basic signal[6]. Similarly shape and dynamic feature are extracted in some cases [7]. The discrete cosine transform is used to analyze the feature and reduce the dimensionality, redundancy of the signature. Divide by zero error and irrelevance of

Signature verification

The signature verification is identified by the biological characteristics of the human. The biological characteristic classified in terms of forgeries. The forgery is imitated by the forger.

Types of forgery

Forgery classified based on the skills of the forger.

  • (i)

    Random forgery

    Forgery person does not have any information about the original person name and signature. These forgeries are easily identified and verified [10].

  • (ii)

    Simple forgery

    The forgery person known the genuine person name and signature.

Proposed feature extraction technique

For signature verification, different features are extracted and verified with different standard databases. Due to many number of feature extraction, the computation time is increased drastically. So delay will be created unnecessarily in forgery detection. In this paper, minimum number of features are selected which will be common to all signature verification and produced accurate authentication result for signature holder. The Fig. 1 shows the proposed feature extraction.

The above block

Performance analysis

Here we took 84 user samples and trained with this extracted features in Neural network. This system detects 95% signatures correctly are shown in Fig. 8a, b, c, d, e, f and g.

Query signature is not in database then it trained first and then getting into test the signature verification.

FAR – False Acceptance Ratio The false acceptance rate is given by the number of incorrect signatures generated by the system compared to the total number of acceptances.FAR=(NFA/NIA)*100%

Where, (NFA=Number of

Conclusion

This paper presents a method of signature verification using IWT and back propagation neural network approach. This method extract the features from preprocessed signature images. The extracted features are used to train a neural network using back propagation training algorithm. As shown in Table 1 tolerance level exceeds than detect the forgery signal. The network could classify all genuine and forged signatures correctly. When the network was presented with signature samples from database

Declaration of Competing Interest

The method uses features extracted from preprocessed signature images. The extracted features are used to train a neural network using back propagation training algorithm. As shown in Table 1 tolerance level exceeds than detect the forgery signal. The network could classify all genuine and forged signatures correctly.

K. Tamilarasi received M.E degree in Anna university in the year of 2009. Currently she is working as Assistant professor in Excel Engineering College, Namakkal. She has published 3 papers in various journals. Her research interest is machine learning and image processing.

References (20)

There are more references available in the full text version of this article.

Cited by (7)

  • An easy-to-explain decision support framework for forensic analysis of dynamic signatures

    2021, Forensic Science International: Digital Investigation
    Citation Excerpt :

    Mavrogiorgou (2001) conducted a kinematic analysis of handwriting movements showing handwriting impairments in patients with obsessive-compulsive disorder. With the advent of deep learning, the latest studies on signature recognition often make use of neural networks to build predictive models (Mersa et al., 2019; Shariatmadari et al., 2019; Vorugunti et al., 2020b; Tamilarasi and Kalyani, 2020). However, these methods, while known to be very accurate, have a reputation for being very complex to explain so that they are ultimately used as black boxes.

  • HSCM: A New Framework For Handwritten Signature Verification Using ConvMixer

    2023, 5th Novel Intelligent and Leading Emerging Sciences Conference, NILES 2023 - Proceedings
  • Dynamic signature verification method based on Pearson correlation coefficient

    2022, Yi Qi Yi Biao Xue Bao/Chinese Journal of Scientific Instrument
  • A Unique Approach to Efficient Fraudulent Signature Detection Using Deep Convolutional Neural Network, Xception, and EfficientNet

    2022, Proceedings - 2022 IEEE International Conference on Artificial Intelligence and Virtual Reality, AIVR 2022
View all citing articles on Scopus

K. Tamilarasi received M.E degree in Anna university in the year of 2009. Currently she is working as Assistant professor in Excel Engineering College, Namakkal. She has published 3 papers in various journals. Her research interest is machine learning and image processing.

Dr. S. Nithya kalyani received B.E degree and M.E degree in Anna University in the year of 2004 and 2006. And also she obtained P.hd Degree in the year of 2014. She guided 14 UG projects, 15 PG project and 8 Research Projects. She has published 35 papers in various journals. Now she is working as Associate professor in the Department of information Technology in K.S.R. College of Engineering, Namakkal. Her research interest is Data mining, Wireless Sensor Network, Computer network and software Engineering.

View full text