Assessing the common authorship of a set of questioned signature images

Abstract

This paper assesses the common authorship of a set of questioned signature images. Our proposal consists in generating as many models as available questioned signatures. Each model is trained with one questioned signature and artificial duplicates worked out from it. The duplication procedure is based on neurocognitive distortions which are inspired from human writing processes. Once the similarity of each signature is evaluated with each model, a similarity score matrix is calculated. The cohesion of this matrix is used to assess whether all specimens were produced either by the same or by more than one writer.