Technological Support of Forensic Examination of Handwritten Material with Introduction of Neural Networks

The article deals with the issues of automation and computerization of forensic handwriting examination and handwriting studies with regard to the introduction of new information technologies – the artificial neural networks – into the expert production. The main purpose of the scientific work is to develop handwriting examination methods that help investigators and judges recognize illegible handwriting in case materials. It substantiates the idea of setting up an automated information retrieval system (IRS) for handwriting examination objects containing handwriting reference samples of those persons who passed the Unified State Exam. The advantages and disadvantages of such handwriting samples are discussed in terms of their completeness, reliability, sufficiency and comparability. Necessary management and legislative decisions are proposed to be taken in order to set up the relevant repository. In addition to providing handwriting reference samples for comparative analysis in the framework of handwriting examination, the handwriting samples repository solves another important task – it contains big data allowing for neural networks training which, in its turn, is capable of solving various identification and diagnostic tasks of forensic handwriting examination. The training data are subject to markup and systematization, and the article suggests the basis for their markup. An algorithm for two-stage recognition of illegible handwriting by neural networks is proposed: the one based on the graphical characteristics of the manuscript and the other one related to morphological and syntactic features of the written speech, which is based on computer vision technologies and the language model of the Russian language. The author presents his own systems of diagnostic and identification features and rules for their calculation, the application of which contributes to solving the tasks assigned to experts.