The Group has a long history of application-led work in the automated analysis and recognition of handwritten data. We have worked extensively in the development of algorithms for handwriting recognition and text processing. This has led us to work more generally in document analysis, typified by a project which addressed some fundamental issues in bank cheque processing, while we have also researched analytical techniques which can help to identify the writer of a handwritten sample, and have concentrated particularly on the practically important scenario where only short written samples are available for inspection.
Similarly, we have long been recognised for our work in automatic signature verification. We have developed architectures to support efficient verification processing, and have looked at both global and individually-oriented feature selection mechanisms to improve achievable performance. As a behavioural biometric, signature verification is especially vulnerable to the existence of individuals with an inherently high degree of unpredictability and variability in their signature samples (so-called "goats"). We have compiled a large database of signature samples from a cross-section of the general public, where all samples were collected in a typical external operating environment. This has allowed us to study the characteristics of a large population of signers, and we have recently been able to carry out a study which has given us new insights into how best to identify and manage the inclusion of goats in a general population.
We have undertaken some very important work on the behavioural characteristics of human observers in signature verification tasks. We have investigated concepts such as "complexity" and "stability" in signature formation, and have studied the relationship between measures which characterise these ideas and the "forgability" of signatures. This work has an intrinsic value in supporting humans working in typical current operational environments, but also has a value in promoting the design of better automated systems.
We are currently developing our work on handwriting analysis in a project which is exploring forensic applications of automated techniques. Working with professional document examiners we are investigating ways in which features which assist in forensic handwriting analysis can be extracted from text fragments. A particular issue which we aim to address is the extent to which the dynamics of the writing process can be accurately inferred from a static writing sample. This is a common requirement in forensic analysis, and we aim to evaluate rigorously the effectiveness and reliability of this process. Our work will also provide an opportunity to develop and evaluate tools which can assist in the training of forensic document analysts.
This research theme has been supported over a number of years by EPSRC (especially some of the work on signature verification, and the current project on forensic document analysis tools), the EU (signature verification), and has close links with our work on the development of generic classification tools, and with our work on biometrics.