Pre-expert verification of forged signatures: human and machine approaches

Abstract

The paper provides a general description of expert and pre-expert; offline and online situations of signature verification as the main requisite of legally significant documents. Examples of cases where such verification may be required, particular features of response to such incidents are given, statistical data on the probabilities of errors by a person and an intelligent system are disclosed. An experiment is described to determine the accuracy of recognition of a forged signature by a person and an artificial intelligence system, during which the average accuracy of verification of handwritten signatures by a person was established, which amounted to 61.18%. The ratio of false-positive and false-negative errors in the verification of signatures by both a human and an intelligent system is given. Separate elements of the NSP-SigVer scientific project are disclosed, within the framework of which an intelligent signature verification system is being developed: hypothesis setting, dataset content, statistical results of experiments with artificial neural networks technology. The article also describes the logic of evaluating the probability of authenticity of a signature offered by an intelligent system.