Bayesian multivariate models for case assessment in dynamic signature cases

Dynamic feature analysis of handwritten electronic signatures based on Fourier transform

The utilization of handwritten electronic signatures has expanded in various application scenarios, leading to an increased demand for identification. Unlike handwriting signatures, handwritten electronic signatures offer the advantage of extracting dynamic feature data, including writing pressure, velocity, and acceleration. In this study, the Fourier transform was employed to extract 18 characteristics from the time domain and frequency domain of writing pressure, velocity, and acceleration. The experimental findings revealed distinguishable differences between genuine signatures and random forgeries in writing pressure. However, no statistically significant differences were observed in writing velocity and writing acceleration. Moreover, significant differences were detected in most characteristics when comparing genuine signatures with freehand imitation forgeries and tracing imitation forgeries. The canonical discriminant analysis was performed between the genuine and Non-genuine signatures; the cross-validation estimated the discriminating power of these characteristics with a satisfactory result. The study proposed a new approach to analyzing handwritten electronic signatures using time-domain and frequency-domain characteristics and demonstrated its effectiveness in the examination.

Bayesian evaluation of dynamic signatures in operational conditions

Forensic handwriting examiners (FHE) activities are focused on comparative analysis of handwritten objects such as signatures. Their role is to provide and evaluate evidence for and against the authenticity of a questioned signature. In recent years, cases involving handwritten signatures captured on electronic devices have become more commonplace. These so-called 'dynamic signatures' (also known as 'digitally captured signatures') are much different from paper-based signatures. Not only does the medium of recording differ, but also the type, volume of data and features are different from the pattern-based evidence that makes up paper-based signatures. Recent developments in forensic science - including signature examination - have led to the adoption of evaluative probabilistic methodologies in many disciplines [see, e.g. ENFSI 1915 Guidelines]. In the current paper, a probabilistic model to evaluate signature evidence in the form of multivariate data, as proposed and described in Wacom Europe GmbH (2019), is adopted. Topics like data sparsity, joint evaluation of multiple features and feature selection are investigated. Performed experimental studies showed an accuracy rate above 90% even when a limited number (5) of reference signatures was available. The performances of a multivariate approach are compared with those characterizing a so-called multiplicative approach where variables (features) are taken as independent and the Bayes' factor (BF) is obtained as the product of univariate BFs associated to each selected feature. The simplicity of this latter approach is, however, accompanied by severe issues about the reliability of results. The use of a multivariate approach is therefore highly recommended. Finally, the evidential values in correspondence of alternative feature sets are compared. Results suggest that discriminative features are writer-related and necessitate a case-specific selection.

Condom evidence: Characterisation, discrimination and classification of pyrolysis-GC-MS profiles

Analysis of condom evidence commonly focusses on the detection of silicone-based lubricants, such as polydimethylsiloxane (PDMS). Although various instruments are used to analyse silicone lubricants, pyrolysis-gas chromatography coupled to mass spectrometry (py-GC-MS) is one of the few instruments that presents immediate applicability to casework. However, considering that this technique detects silicone-based evidence, it is important to evaluate the discrimination potential of the method when applied to various samples. Examination of the variability within a large sample set from an international market is needed to evaluate the most distinguishing compounds likely to be detected in casework. In this study, 70 condoms, personal hygiene products, and lubricants, were analysed using py-GC-MS. Resulting pyrograms were characterised using published spectral databases. Pyrolysates data were extracted and evaluated using multivariate techniques. Qualitative visual inspection of the data, as well as statistical analysis, revealed at least five groups within the dataset. Discrimination was based on four main oligomers, as well as six minor compounds from siloxane degradation. Condom lubricants were found to present a very regular pattern, allowing for the majority of them to be differentiated from personal lubricants.