Journal of Global Research in Computer Sciences

Journal of Global Research in Computer Sciences

Menu

ISSN: 2229-371X

All submissions of the EM system will be redirected to Online Manuscript Submission System. Authors are requested to submit articles directly to Online Manuscript Submission System of respective journal.

HANDWRITTEN SIGNATURE VERIFICATIONS USING ADAPTIVE RESONANCE THEORY TYPE-2 (ART-2) NET

Tirtharaj Dash*1, Subhagata Chattopadhyay2 and Tanistha Nayak3
  1. Department of Computer Science & Engineering Veer Surendra Sai University of Technology Burla-768018, India
  2. Department of Computer Science & Engineering Bankura Unnayari Institute of Engineering Bankura-722146, India
  3. School of Computing National Institute of Science and Technology Berhampur-761008, India
Corresponding Author: Tirtharaj Dash, E-mail: tirtharajnist446@gmail.com
Related article at Pubmed, Scholar Google

Visit for more related articles at Journal of Global Research in Computer Sciences

Journal of Global Research in Computer Sciences

Abstract

Authorizing hand-written signature has always been a challenge to prevent illegal transactions, especially when the forged and the original signatures are very „similar-looking? in nature. In this paper, we aim to automate forged signature verification process, offline, using Adaptive Resonance Theory type-2 (ART-2), which has been implemented in „C? language using both sequential and parallel programming. The said network has been trained with the original signature and tested with twelve very similar-looking but forged signatures. The mismatch threshold is set as 5%; however, it is set flexible as per the requirement from case-to-case. In order to obtain the desired result, the vigilance parameter (ρ) and the cluster size (m) has been tuned by carefully conducted parametric studies. The accuracy of the ART-2 net has been computed as almost 100% with ρ = 0.97 and m = 20.

Keywords
handwritten signature; automatic verification; ART-2; forged signatures

INTRODUCTION
Neural networks have been widely used in pattern recognition, especially where the patterns are complex due to close resemblance of „original‟ and „generated‟ patterns [1-4]. An important property of a neural network classifier is that, it learns the exemplary patterns (as inputs) by updating its nodal connectors‟ weights. The drawback of such type of learning is that, when new patterns are fed, the weights are updated and as a result, it loses the memory of older patterns and stores the impression of new patterns [5]. To handle this issue, Grossberg and Carpenter (1987) proposed the concept of Adaptive Resonance theory (ART) networks, where the networks retain the earlier learning, which is certainly advantageous over the conventional neural classifier [6].
ART is of two types i.e. type-1 and type-2. ART-1 takes binary input vector, whereas, ART-2 takes analog/continuous input vector and therefore more meritorious [7]. In our earlier work, ART-1 network has been considered for automatic verification of hand-written signature offline, with high level of accuracy (99.97%) [8]. In that paper, however, only two forged signatures were considered. In this paper, ART-2 has been considered for the offline verification of twelve very similar looking but forged handwritten signatures.
Handwritten signature is the principal biometric measure for personal identification. It is an important method for performing legal transactions. However, there are chances when signatures could be delicately copied, such that these apparently resemble originals and are difficult to be identified by the naked eyes. Hence, automating such a detection process could be of real advantage to us. However, it requires vast research prior its practical use. In this view, this paper is an attempt where ART-2 net has been used and implemented using both sequential and parallel programming techniques to note its detection accuracy and speed.
Automatic verification of handwritten signature is an age-old research topic. Available literatures show that several traditional and soft computing techniques have been used for accomplishing the said task. Due to space constraints, detail discussion of all the techniques are beyond the scope of this paper. Hence, some relevant studies have been shown, below.
image
image
From these studies, it may be noted that ART has not been tested widely in this field, which leaves an opportunity to investigate ART-2, which is the motivation behind this work.
In the following section, we have described the methodology of ART-2 implementation using „C‟ language using both the sequential and parallel programming.

METHODOLOGY
image
image
image
image

RESULTS
The average similarity index (SI) between the original and forged signatures near 51%, which may have higher chance of matching, instead of rejecting the forged signatures. It is desired that even with slightest difference, the network must be able to differentiate those from the original signature based on its learning and assigned vigilance. The paper suggests that vigilance parameter (ρ) needs to be optimally set, which is the first challenge. In this work, optimum ρ has been set through a detail parametric study (see table-1 and 2). The second challenge is to assure that the network learns the exemplary patterns through several observations (number of clusters).
Table 3 shows how the cluster size (m) influences the accuracy and the computational times. In table 1, it may be seen that with forged signatures 11 and 12 the mismatch is <5% and therefore these are accepted as original. In case, the mismatch threshold is set <1%, the algorithm would be able to detect all forged signatures. Hence, we have made the algorithm very flexible to allow such modification, which depends on the situations. Table 2 shows that with ρ=0.97, the detection accuracy is almost 100% with minimum time in both sequential and parallel programming.
image
image
Fig.3 plots the „ρ vs. accuracy‟ parametric study. As seen in table 2, for ρ = 0.97, the accuracies are 99.9989 in both the sequential and parallel programming, we have shown the plot for parallel processing.
image
image
Table-3 shows that with „ρ=0.97‟ and „m‟ = 20, the detection accuracy is the highest. Fig. 4 plots the number of clusters (m) vs. the respective accuracy levels achieved.
image
Therefore, we conclude that through parametric studies, our method gives more accurate result when compared with other techniques, described in section I.

CONCLUSIONS AND FUTURE WORK
An ART-2 type net has been developed in this work for automating the verification of very similar looking (SI ~51%) forged signatures, offline. It has been implemented with both sequential and parallel processing to achieve faster and accurate detection with a mismatch threshold of 5%. Through parametric studies the best „ρ‟ and „m‟ are obtained. The accuracy is found to be 99.98%.
It is important to mention that, in this study we have tested only twelve forged signatures, which is a small sized sample. This is certainly a limitation of this work. The net needs to be tested with many different types of original as well as forged signatures for obtaining a more authentic proof of its performance. We are currently working on it.

References
  1. S. Chattopadhyay, P. Kaur, F.A. Rabhi, U.R. Acharya– “Neural Network Approaches to Grade Adult Depression”, Journal of Medical Systems (2011), published online on 21/07/2011 DOI: 10.1007/s10916-011-9759-1
  2. S. Chattopadhyay, P. Kaur, F.A. Rabhi, U.R. Acharya – “An Automated System to Diagnose the Severity of Adult Depression”. In the proceedings of 2nd International Conference on Emerging Applications of Information Technology (CSI EAIT-2011), 19-20th February Kolkata India, (Editors - Jana D., and Pal P.), pp. 121-124. Publishers: IEEE Computer Society and Conference Publishing Services IEEE Xplore, Google Scholar, DOI: 10.1109/EAIT.2011.17.
  3. Chattopadhyay S., Pratihar D. K, De Sarkar S. C. –"Statistical Modelling of Psychoses Data”. Computer Methods and Programs in Biomedicine (2010); 100(3): pp. 222-236.
  4. S. Chattopadhyay., D. K. Pratihar, S.C. De Sarkar - "Some Studies on Fuzzy clustering of psychosis data". International Journal of Business Intelligence and Data Mining (2007); 2(2): pp. 143-159.
  5. S.K. Ahmed, A.K. Ramasamy, A. Khairuddin, J. Omar. “Automatic online signature verification: A prototype using neural networks”. TENCON 2009 pp. 1-4.
  6. G.A. Carpenter, S. Grossberg, “A self organizing neural network for supervised learning, recognition, and prediction. Can neural networks learn to recognize new objects without forgetting familiar ones?” IEEE Communication Magazine, 1992, pp. 38-49.
  7. S.N. Sivanandan, S.N. Deepa “Principle of Soft Computing (2nd Edition)” WILEY-INDIA, 2011. ISBN-978-81-265-2741-0, 2011
  8. Dash T., Nayak T., Chattopadhyay S., “Offline Verification of Hand Written Signature Using Adaptive Resonance Theory Net (Type-1),” ICECT-2012, Vol-2, pp. 205-210, 2012.
  9. S. Inglis, I.H. Witten. “Compression-based Template Matching”. Proc. IEEE Data Compression Conference. Los Alamitos, CA, pp. 106-115, 1994.
  10. Y. Mizukami, M. Yoshimura, H. Miike, I. Yoshimura. “An off-line signature verification system using an extracted displacement function”. Pattern Recognition Letters. 23, 1569-1577, 2002.
  11. A. Kholmatov, B. Yanikoglu. “Identity authentication using improved online signature verification method”. Pattern Recognition Letters. 26, 1400-2408, 2005.
  12. A.P. Shanker, A.N. Rajagopalan. “Off-line signature verification using DTW”. Pattern Recognition Letters. 28, 1407-1414, 2007.
  13. W. Tian, Y. Qiao, Z. Ma. “A New Scheme for Off-line Signature Verification Using DWT and Fuzzy Net”. ACIS Int. Conf. Software Engineering, Artificial Intelligence, Networking, and Parallel/Distributed Computing,. Vol.3, pp.30-35, 2007.
  14. K.R Radhika, M.K. Venkatesha, G.N. Sekhar. “An approach for online signature authentication using Zernike moments”. Pattern Recognition Letters. 32, 749-760, 2011.
  15. A.K.M. A., Hossain, Md. G. Rahman, R.C. Debnath. “Comparative study of signature verification system using Supervised (BPNN) and Unsupervised (HMM) Models”. ICCIT-2003, pp. 732-737, , 2003.
  16. H. Lv, W. Wang, C. Wang, Q. Zhuo. “Off-line Chinese signature verification based on support vector machines”. Pattern Recognition letters. 26, 2390-2399, 2005.
  17. M.S.Aksoy, H. Mathkour. “Signature verification using rules 3-ext inductive learning system”. Int. J. Physical Sciences. 6(18):4428-4434, 2011.
  18. Dash T., Nayak T., Chattopadhyay S., “Offline Hand Written Signature Verification Using Associative Memory Net ”. Int. J. of Advanced Research in Computer Engineering & Technology. 1(4):370-374,2012.
  19. S. Chattopadhyay, S. Banerjee., F.A, Rabhi, U.R.Acharya “ A Case-based Reasoning System for Complex Medical Diagnoses.” Expert System :The Journal of Knowledge Engineering (2012); DOI:10.1111/j.1468-0394.2012.00618.x (in press).