Performance Evaluation ofSignature Verification System Using Deep Learning Model forOffline Signatures

  • Prof. Minal D. Shahakar, Ayushi S. Tiwari, Monica M. Baloji


Abstract: Handwritten signatures are generally acknowledged as a method of document authentication, authorization and individual confirmation.There are two kinds of signatures, offline (static) and online (dynamic). Offline signatures are increasingly hard to verify. The drawback of offline signatures is that they cannot be signed in the similar way even by the owner of the signature. This is called intra-personal fluctuation. All these make the offline signature verification a challenging issue for specialists. This paper proposed a Deep Learning (DL) based offline signature verification technique to forestall signature fraud by malignant individuals. It presents the Convolutional Neural Network (CNN) algorithm for feature extraction for the check of offline signatures. The cropped signatures are utilized to train CNN for extracting features.CNN is designed and trained for thresholding. While classifying whether a given signature was a fraud or authentic, the test results indicate that it accomplished standard accuracies. In this paper CNN model was made utilizing python for offline signature and after training and approving, the exactness of scanned signatures of testing was 95%. It is anticipated that the success of the acquired outcomes will increment if the CNN method is upheld by including additional feature extraction techniques.