Handwritten Signature Recognition using Neural Networks

This work is devoted to solving the problem of signature recognition using neural networks. The authors proposed the use of convolutional neural networks to determine the signature class. Signatures play an important role in financial, commercial and legal transactions, their recognition guarantees the security of not only information, but the whole person. The use of neural networks for signature recognition allows you to reliably identify a user in an automated mode. The authors have developed a convolutional neural network, and also proposed an algorithm that consists of image preprocessing, including background segmentation, noise reduction, and image normalization. Image preprocessing improves the quality of the network. Next, feature extraction is performed, which consists of global features, such as the ratio of the height to width of the signature, the maximum horizontal histogram and the maximum vertical histogram, the horizontal center and the vertical center of the signature, the endpoints of the signature, the signature area, training a neural network with extracted features, recognition the owner of the handwritten signature and then predicting the class of the signature.

Purpose – development of a handwritten signature recognition algorithm using neural networks.

Methodology: the methods of computer vision were used in the work; deep learning methods, as well as object-oriented programming methods.

Results: developed a handwritten signature recognition algorithm using a neural network.

Practical implications: the application of the results obtained is useful in forensic analysis of documents, since a person uses a signature on a regular basis to sign checks, legal documents, contracts and other paper media that need protection. Therefore, when someone tries to copy a signature, a problem arises that can lead to undesirable consequences in the form of theft and further use of both personal data and other valuable secret information.