An Algorithm for Classifying Handwritten Signatures Using Convolutional Networks

Abstract

In this study, a model based on convolutional neural networks is proposed to quickly and efficiently classify and identify a person's signature with over 90% accuracy. For this purpose, two signature datasets were used. The first, called CEDAR, is publicly available. The second set, called GC-DB, was collected by the researchers using uncontrolled environments (different signing positions). This dataset has 121 local signatories from the Republic of Ecuador, who submitted 45 copies of signatures each. In this set of signatures, the implicit noise produced by the capture device and by the paper of different thicknesses used in the collection made noise removal a relatively complex operation. The effectiveness of the proposed algorithm was compared with two other algorithms that were implemented and validated using the two data sets. The results show that it is possible to perform an efficient classification of handwritten signatures with the developed algorithm. Additionally, the developed algorithm is lightweight and easy to implement and can be installed on portable devices such as cell phones or tablets.