Automated Verification of the Signatures' Authenticity Using Artificial Intelligence Methods
Piotr Bilski, Jacek Olejnik, Mieczysław Goc · 2023
The paper presents the approach to identify fake signatures based on the image analysis. The problem is related with the forensics operations in order to distinguish handwritten signatures made by the human from the machine-originated counterfeits. The identification system is based on the selected artificial intelligence-based classifiers and processes features extracted from the signature images. The source material comes from either human or the 5-dimensional printer, being able to mimic a hand with the pen writing on the sheet of paper. Each image is then processed by the profilometer to extract important information allowing for distinguishing the original signature from the false one. Features include attributes related with the pen's position and inclination, obtained through the Fourier and wavelet transformations. The identification is then made by the intelligent classifier. For the project multiple algorithms were selected and tested regarding their accuracy, such as multilayered perceptron, decision tree or k Nearest Neighbors classifier. Presented experimental results show the ability of the system to support human during the task of the signature authenticity verification.