Automatic Handwriting Analysis to Support and Evaluate Document Inspection in Criminal Investigations
Handwriting identification is a widely used forensic technique which can play a key role in providing evidence in fraud, identity theft, and even murder cases. Handwriting is usually evaluated by human visual inspection but this process can be prone to a degree of subjectivity. Although some computer-based tools are available to assist the examination process, the accuracy of assisted interpretation still depends principally on human skill. This EPSRC-supported project being undertaken in collaboration with professional document examiners from Document Evidence Ltd, is seeking to exploit automated handwriting analysis technology to address some key issues in forensic handwritten document examination which will improve our understanding of these processes, and the way in which document examiners’ skills can be developed.
The research will develop methods to compare and validate a forensic document examiner’s analysis of the structure of a handwriting sample and internal characteristics against measurable features extracted using computer-based analysis. This will encompass the direct comparison of "static" features of the writing (i.e. features related to the 2D image generated) and also the predictive accuracy, compared to direct computer-based measurement, of "dynamic" features of the writing execution embodied in samples prepared through on-line writing capture.
An important aspect of the project is to compare the relationship between features which have to be inferred by human inspection of handwriting samples and accurate measurements which can be made automatically when on-line capture of samples is possible. This is illustrated in the accompanying Figure which shows a small section of segmented handwritten text. A document inspector might well estimate pen trajectory by an evaluation of relative (predicted) pen pressure within the image (Regions A and B, for example, show evidence of this). The dynamic pen path reconstructed in the lower trace (as a result of on-line capture) allows the pen-ups to be measured exactly while, indeed, the whole pressure profile during execution can also be recorded (bottom right) for further analysis.
This sort of analysis will provide the basis of an evaluation of current document inspection protocols, the development of new analytical tools, and the implementation of aids for the effective training of document examiners. We are currently embarking on a major handwriting collection trial to provide targeted data for our experimentation.