Handwritten Signature Verification using TensorFlow

Abstract

Many different methods are being tested by researchers for the correct identification of original signatures from forged ones. The objective of the research reported in this paper is to set up a new system that can be used reliably for the recognition of signatures in legal documents and cheques used in banks. The proposed system was designed using TensorFlow, which is used widely for deep learning. The Convolutional Neural Network (CNN) used in the designed system is capable of accurately verifying the characters unique to the original signature. The effectiveness of the system is measured using two parameters which are False Rejection Rates (FRR) and False Acceptance Rate (FAR). The proposed system showed FAR and FRR values as 5 percent and 5 percent respectively while testing and the overall accuracy of the system is 90 percent.