Expertise in Applied Face Matching: Training, Forensic Examiners, Super Matchers and Algorithms

Face matching is widely used in applied settings, including policing and border control, to identify persons of interest, where the consequences of an incorrect decision can have profound consequences. It is, therefore, of paramount importance that applied face- matching systems are accurate and reliable. However, humans are generally poor at matching face of people they don’t know, with large individual differences in accuracy. The aim of this thesis was to evaluate different sources of face-matching expertise (training, forensic face examination, superior face matchers and algorithms) and provide recommendations for how to improve face-matching performance in applied settings. Study one presents a survey of face-matching training, providing insights into the diverse and inconsistent approaches organisations use to train face-matching operators. The second study evaluates a two-day professional face-matching training course, demonstrating the limitations of short courses and the risk of introducing a match bias in low performers. In study three the perceptual skill of superior face matchers and forensic face examiners were compared, showing that by combining the selection of high performers with a wisdom of crowds approach, comparable levels of performance to trained examiners can be achieved in quick-decision face matching. Study four investigated the fusion of human face-matching decisions and algorithm similarity scores for faces that were challenging to humans and to the algorithm, highlighting the effectiveness of fusion in improving face-matching performance. Study five compared the operational accuracy of individual examiners and examiner teams on a face-matching task. Teams achieved higher levels of performance than individuals, with performance improving for both groups after fusion with a facial recognition algorithm. The thesis concludes with a discussion of how different sources of face-matching expertise can be used and combined in applied face-matching systems, and highlights areas for further research that would benefit the applied face-matching community.