- Prof. Nick Fieller – Probability & Statistics – SoMaS, University of Sheffield
- FCUL (DEIO) – Campo Grande – Bloco C6 Piso 4 – Sala 6.4.31 – 14:30h
- Quarta-feira, 11 de Maio de 2011
Abstract: Fingerprints have been used to provide a unique identification of individuals in criminal cases for 120 years. Computer based automatic systems were first introduced 40 years ago. However, despite this long history techniques for searching for matches are based only on heuristics and frequently the best 10 or more matches found are eliminated by other evidence. Even with the best systems the final agreed match is found in the third or fourth set of ten presented. The ultimate Yes/No judgment is made by an expert using “professional experience”. Often experts disagree and there are an increasing number of proven cases of incorrect matching of fingerprints by experts.
Recently the UK Forensic Science Service has commissioned a study aimed at providing a statistical approach to measuring the quality of a tentative match using likelihood ratios to provide a probabilistic measure. This follows the procedure used in the evaluation of other types of forensic evidence (glass and fibres, faces etc). This talk describes some of the early work on this study, starting with a brief historical review and description of the macro and micro features of fingerprints, why these might be modelled by random processes and how this might be utilised to provide a full likelihood based evaluation of matches.