Offline Signature Verification: A Study on Total Variation versus CNN

Abstract

This paper studies offline handwritten signature verification and the authenticity of a given signature. The research in this paper develops and compares two algorithms that predict forgery and authentic signatures based on the acquired set of images. For the first method, we use total variation technique as a measure of contiguity in the signatures to test its ability to verify the genuineness of a signature. Convolutional Neural Networks (CNN) was chosen as a second approach for signature validation. CNN is a powerful class of deep learning architecture. The algorithms described in the paper have been proven to be low cost as well as to make predictions with high accuracy in handwritten signature authentication.