Generative‐adversarial network for falsification of handwritten signatures

With further development of generative AI, primarily generative-adversarial networks (GAN), deepfakes are gaining in quality and accessibility. While, forensic methods designed for examination of handwriting are often applied to its digital copies, despite being possibly insensitive to cases of GAN-made forgeries (unless methods of digital forensics are co-employed). Approaching this problem from a novel perspective, we have created a translational GAN tasked with generating false handwritten signatures from limited examples, aiming to ascertain whether traditional methods of signature examination will be effective against such forgeries. We have found that traditional methods of handwriting examination are sufficient for identification of discriminative features that could result in rejection of GAN-made forgeries, however, those stemmed mostly from the lesser visual quality of the generated signatures, which could be improved in the future.