Verification of a static (off-line) signature using a convolutional neural network

Abstract

This article is devoted to the development of a method for detecting forgery of handwritten signatures. The signature still remains one of the most common methods of identification. The signature on financial and other documents can be forged, so detecting forgery is an urgent task. This is the task of binary classification: to determine whether the signature is genuine or fake.The article describes the results of recognition of handwritten signatures made on paper. A database of handwritten signatures of 10 people was used for experiments. For each person, 10 genuine and 10 forgery signatures made by other people were collected. The signatures were digitized as color images with a resolution of 850×550 pixels. Then a binary representation of each signature was formed. Three variants of reducing signatures to sizes were used for classification: 128×128, 256×256 and 512×512 pixels. These images served as the source data for the convolutional neural network.As a result of testing the proposed approach, the average accuracy of the correct classification was achieved on medium-sized images and is equal to 93.33%.