Offline Handwritten Signature Recognition Using Deep Convolution Neural Network

Abstract

In the modern age, technological advancement reached a new limit where authentication plays a vital role in security management. Biometric-based authentication is the most referenced procedure for authentication where signature verification is a significant part of it for authentication of a person. To prevent the falsification of signatures on important documents & legal transactions it is necessary to recognize a person's signature accurately. This paper focused on recognizing offline handwritten original & forged signatures using a deep convolution neural network. We use a completely new dataset & also downloaded datasets to train the system & verify a random signature as genuine or forgery. All testing samples are collected from several individuals after several steps of preprocessing the model is fed with the resultant image to our system, the experimental results give us an accuracy of 95.5% from the dataset.

Offline signature verification using deep learning method

“…In a skilled forgery case, the forger knows both the signatory's name and signature shape with some practice to make a good faking of the signatory's signature. It is difficult to verify this kind of forgery because of its high similarity to the signatory's signature [3,4]. Figure 1 illustrates some samples of genuine signatures for some writers and how the skilled forged signatures for them are similar to the genuine signatures.…”