Skip to main content

Bayesian Biometrics For Forensics

Final Report Summary - BBFOR2 (Bayesian Biometrics For Forensics)

In the EU 7th framework programme Marie Curie Initial Training Network “Bayesian Biometrics for Forensics” (BBfor2), eighteen young researchers have received training in the multidisciplinary area of biometric technologies in forensic science through conducting scientific research. Most of these Marie Curie Fellows have carried out their work in the light of a PhD thesis, and three of the fellows functioned as early Postdoctoral researchers. The biometric technologies covered automatic speaker recognition, automatic face recognition, and automatic fingerprint and palmprint recognition. In all of these, the forensic application was always central: the condition where the quality of the sample (e.g. the speech recorded on a voice-mail or the image of the face on a security camera footage) is low when compared to commercial security applications. For instance, the fingermark left behind on a glass in a crime scene is less complete, and can be smudged and be somewhat deteriorated, contrary to the fingerprint acquired by the home button on a modern iPhone will be in pristine condition.
The fellows have worked on improving the quality of the basic biometric detectors by implementing better feature extraction and by investigating compensation schemes at the levels of feature, model and comparison scores.

All of the technologies worked together in a larger framework, investigating the question on how a biometric comparison (a “match” in popular parlance) can be presented in court in terms of a weight-of-evidence. The notion of the computation of a likelihood ratio---comparing the likelihoods of the prosecutor’s hypothesis (the suspect is the culprit) to likelihood of the defence hypothesis (someone else did it)---has been picked up in all the biometric technologies covered in BBfor2. Not only did we standardize the use of the Likelihood Ratio throughout these disciplines, we also homogenized the paradigm to evaluate the quality of the reported likelihood ratios. In this approach, we can validate the application of a biometric technology in forensic science by determining the discrimination ability of the detector independently from its ability to produce meaningful likelihood ratios. We call the latter ability calibration, and it is important that a biometric comparison gives well-calibrated likelihood ratios, because if this is the case, the value can be used directly in a court case as the weight of evidence. Such a calibrated comparison from an automatic recognition system has many advantage. A system is objective, it is always available, and the contributions from independent pieces of evidence can easily be combined. The progress made in this area will have a direct effect on how the evidence in court will be presented in the future, and with the finalization of BBfor2 there are young professionals ready to translate the knowledge acquired in the project to the various areas in the legal system in Europe and beyond. Hence we think the results of the work are very relevant to policy makers and professionals active in the police and justice system in Europe.

The fellows did not carry out their work in isolation, in contrary, the MC ITN framework provided many opportunities to learn from other fellows and supervisors in the Network. All fellows have carried out at least two secondments at other host institutions in the Network, which in many cases has led to joint publications. Often these secondments were truly interdisciplinary, e.g. a speech researcher visited a face recognition group in a different research institute. These exchanges have turned out extremely fruitful, in some cases new approaches in one research field were successfully introduced which was inspired by earlier successes in another field.

Throughout the project, the Fellows were trained in various scientific aspects across the disciplines through seven workshops and three summer schools organized by BBfor2. They further improved their presentation skills by regularly presenting their progress to the other fellows, which were also good occasions for scientific exchange of ideas and cross pollination of approaches and techniques. At the end of the project, a conference was organized entitled “Biometric Technologies in Forensic Science” where the main results of the project were disseminated to the scientific community and where the interaction with other researchers in the area of biometrics and forensic science was consolidated.

More information can be found on the Project’s web site www.bbfor2.net and on the conference web site http://www.ru.nl/clst/btfs/btfs-2013