Towards Improving Offline Signature Verification Based Authentication Using Machine Learning Classifiers
Kamlesh Kumari1, Sanjeev Rana2

1Kamlesh Kumari ,Research Scholar, Department of Computer Science Engineering , M. M (D.U), Mullana, Ambala.
2Dr. Sanjeev Rana, ,Professor, Department of Computer Science Engineering , M. M (D.U), Mullana, Ambala.
Manuscript received on 25 August 2019. | Revised Manuscript received on 04 September 2019. | Manuscript published on 30 September 2019. | PP: 3393-3401 | Volume-8 Issue-11, September 2019. | Retrieval Number: J99100881019/2019©BEIESP | DOI: 10.35940/ijitee.J9910.0981119
Open Access | Ethics and Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Abstract: Signatures have been accepted in commercial transactions as a method of authentication. Digitizing credentials reduce the storage space requisite for the same information from a few cubic inches to so many bytes on a server. The most frequent use of offline signature authentication is to reduce the turnaround time for cheque clearance. In this paper, machine learning classifiers are used to verify the signature using four image based features. BHsig260 dataset (Bangla and Hindi) has been used. We used signatures of 55 users of Hindi and Bangla each. .Six classifier i.e. Boosted Tree, Random forest classifier (RFC), K-nearest neighbor, Multilayer Perceptron, Support Vector Machine (SVM) and Naive Bayes classifier are used in the work. In the paper, the results of Writer independent model show that accuracy of Hindi off-line signature verification is 72.3 % using MLP with the signature sample size of 20 and that of Bangla is 79 % using RFC with the signature sample size of 23.In user dependent model, for some users, we achieved accuracy of more than 92 % using KNN and SVM.
Keywords: Forensic Handwriting Expert (FHE), UTSig (University of Tehran Persian Signature), K-nearest neighbor (K-NN), Writer Dependent (WD), Writer Independent (WI).
Scope of the Article: Machine Learning