Writer Independent Manipuri Offline Signature Verification using Transfer Learning Technique of Convolutional Neural Network
Teressa Longjam1, Dakshina Ranjan Kisku2
1Teressa Longjam, Department of Computer Science and Engineering, National Institute of Technology, Imphal (Manipur), India.
2Dakshina Ranjan Kisku, Department of Computer Science and Engineering, National Institute of Technology Durgapur (West Bengal), India.
Manuscript received on 22 November 2019 | Revised Manuscript received on 03 December 2019 | Manuscript Published on 14 December 2019 | PP: 76-80 | Volume-9 Issue-1S November 2019 | Retrieval Number: A10181191S19/2019©BEIESP | DOI: 10.35940/ijitee.A1018.1191S19
Open Access | Editorial and Publishing 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: Automatic Signature Verification system is used to verify whether a signature is genuine or forged. Forged Signatures are those signatures that a person produced by imitating the signature of another person. Automatic Signature Verification is very important as a person’s handwritten signature is used everywhere to authenticate themselves and there is not very much difference between a genuine signature and the imitation of it, i.e. a forged signature. In this work, signature verification is done using different pre-trained Convolutional Neural Networks (CNNs). Convolutional Neural Network has powerful learning ability, and it can be used to distinguish between a genuine and a forged signature automatically. In this experiment, Manipuri signature dataset was used, the dataset was prepared originally and it contains 729 genuine signatures and 243 forged signatures. Features were extracted from pre-trained networks and classification was done using binary Support Vector Machine (SVM) classifier and the performances of the networks were compared. And according to the experiment we achieved a classification accuracy of 84.7 using VGG19 features, accuracy of 86.8 using VGG16 features and accuracy of 81.9 using Alexnet features.
Keywords: Forged Signatures, Genuine Signatures, Transfer Learning, Writer Independent Signature Verification.
Scope of the Article: Network Architectures