Bayesian Biometrics for Forensics

With the availability of more advanced biometric technologies the work of forensic sciences is expected to change dramatically in the future. A new elite of researchers will be necessary to work in the fields of investigation and in legal cases in the various European forensic institutes, who will be specialized in extracting and interpreting evidence from biometric traces. Training such an elite group of researchers is the primary goal of the Bayesian Biometrics For Forensics Initial Training Network. We intend to achieve that goal by recruiting a number of Early Stage Researchers (ESRs) in various biometrics labs and providing them with a training programme that will both address the fundamental biometric detection technologies issues and the typical forensic usage and requirements of these biometrics. In recent years, automatic biometric recognition technologies have matured to a level at which they can become useful in the forensic sciences. On the one hand—for the modality of speech, in speaker recognition—systems have shown increased detection performance in yearly NIST benchmark evaluations, showing increasing robustness against the hard problem of channel effects using new compensation approaches. On the other hand, the traditional (semi-automatic) approach in forensic fingerprint matching has undergone a shift from a binary match interpretation towards a likelihood-ratio interpretation of the similarity between evidence sample and suspect sample. The usage of (calibrated) similarity scores makes it possible to use the biometrics for investigation purposes, where the task is to search for suspects in a large collection of model samples. Evaluation of the biometric technology in reference databases makes is possible to determine calibrated likelihood ratios. This way, a Bayesian interpretation of the suspect-evidence similarity can be utilized to combine evidence of different biometric modalities, and eventually provide a numerical value to the evidence in forensic cases. A third field of biometrics, face recognition, is evolving towards forensic application by considering non-frontal and low resolution samples as can be obtained from, e.g., surveillance cameras.

The combination of the work in core biometric detection areas of speaker, fingerprint and face recognition and the experience in forensic research environments will educate these ESRs to become experts in the various biometrics modalities, with the application of forensic research in mind. The multidisciplinary training makes the ESRs very employable in forensic sciences and governmental bodies such as law enforcement institutions, as well as research and development departments of industry.