Static signature verification based on machine learning

Abstract

This paper describes the results of handwritten signature recognition. A handwritten signature database of 40 people made on paper and a publicly available Bengali handwritten signature database of 100 people were used for the experiments. A handwritten signature database of 40 people was collected with 10 authentic and 10 fake signatures for each person made by other people. A Bengali handwritten signature database of 100 people was collected 24 authentic and 30 forged signatures for each person. For this experiment, 20 people were randomly selected from the Bengal Handwritten Signature Database. Four options were used to reduce the signatures to sizes: 200×120, 250×150, 300×150, and 400×200 pixels for classification. These images served as input data for the proposed network architecture. As a result of testing the proposed approach, the average accuracy of correct classification for the first base of handwritten signatures reached 90.04%. For the base of Bengal handwritten signatures 97.50%.