Forensic Verification of Author of Handwritten Character “k” Using Artificial Intelligence

Abstract

Handwriting, as a result of the interplay of physiological and psychological factors, is considered a unique manifestation of human individuality. It carries significant information about the writer, which can be utilized across various fields, particularly in forensic science, where it is applied in forgery detection, psychological profiling, author identification and verification. Forensic handwriting analysis is currently facing a decline of handwriting. Despite the decline, its significance remains relevant as the need for higher-quality expertise is increasing. This leads to an effort to integrate modern technologies within this discipline. A major shortcoming of current studies is that analysis is performed at the sentence or word level. Therefore, a method was proposed for verifying authorship based exclusively on information derived from variations in the handwritten character “k”. This research was conducted in collaboration with the Institute of Criminalistics in Prague and its significance lies in its ability to analyze handwriting at the grapheme level by employing deep learning-based approaches. All models were trained and evaluated on authentic handwriting samples from criminal cases, ensuring practical applicability. The best results were achieved using a siamese network with accuracy of 82.8 %.