School of Human Sciences
Face and Voice Recognition Lab
Institute of Lifecourse Development
University of Greenwich
London
Twitter: @GRecognisers
27 July 2021
Low prevalence match and mismatch detection in simultaneous face matching: Influence of face recognition ability and feature focus guidance
Callan Dray (MSc Psychology: Conversion Graduate, University of Greenwich, 2020)
During 2020, two experiments were conducted to assess the impact of low target prevalence on face matching accuracy. The first experiment was part of Callan Dray’s dissertation for his overall Distinction Level MSc Psychology (Conversion) at the University of Greenwich. This effect was tested for the first time in individuals with superior face recognition ability and the project also assessed the use of internal or external facial feature guidance scales. We would like to thank all participants from the University of Greenwich Volunteer Database who contributed to these experiments.
This project has now been published in the journal Attention, Perception, & Psychophysics and can be found on the link below.
Davis, J. P., Dray, C., Petrov, N., & Belanova, E. (2021). Low prevalence match and mismatch detection in simultaneous face matching: Influence of face recognition ability and feature focus guidance. Attention, Perception, & Psychophysics, 83(7), 2937-2954. https://doi.org/10.3758/s13414-021-02348-4 (download author's free open access link here: https://rdcu.be/ctXOc).
The data for this project are available for download on OSF (https://osf.io/x5s2t/).
Despite being essential in many security and policing operations, simultaneous face matching is highly error prone (See review Robertson et al., 2019). Most past research used trials with an even split of matched and mismatched pairs. However, further reductions in accuracy are observed when target faces are rare (Weatherford et al., 2020). This low prevalence effect may have a strong influence on the performance of staff in workplaces where targets appear infrequently.
Situations include border control when passport officers are tasked with correctly detecting rare fraudulent passports as most documents presented by travellers are legitimate (low prevalence mismatch scenario). However, an opposite pattern is commonly found by operators when using automatic face recognition systems in the street or at events. In these circumstances, depending on the settings, staff must disregard regular alerts of innocent bystanders while ensuring the far rarer alerts for actual suspects are dealt with (low prevalence match scenario).
Therefore, this research aimed to examine the effects of low target prevalence on face matching accuracy. Secondly, we also aimed to observed whether individuals with superior face recognition ability were able to overcome any low prevalence effects and maintain their outstanding face matching accuracy levels (Bobak, Hancock, & Bate, 2016). Finally, it assessed whether the implementation of internal or external facial feature guidance scales improved performance levels (Towler, White, & Kemp, 2017). Participants were assigned to a low prevalence mismatch condition, a low prevalence match condition, or an equal match-mismatch condition.
Experiment 1
In the first experiment super-recognisers (n = 317) significantly outperformed average-ability control participants (n = 452) and an internal feature guidance approach enhanced accuracy across all conditions (albeit effect sizes were far smaller than those for the comparison of the two ability groups). However, control participants, but not super-recognisers, displayed opposite expectations than expected by displaying higher accuracy when under low prevalence conditions. The researchers suspected that these patterns were a result of participants not being informed of their low prevalence conditions prior to the experiment.
Experiment 2
As a result, in a second experiment, above-average range ability participants (n = 841) were provided with information as to their allocated prevalence conditions. By “above average” we mean participants whose face recognition ability is below that of super-recognisers, but above that of Experiment 1’s average-ability controls. This time, the results matched previous research into low prevalence effects.
Overall
Together these results suggest that super-recognisers remained largely unaffected by the low prevalence effect, supporting their deployment within face verification roles. Results also showed small benefits for adopting an internal feature guidance approach in the workplace, although this came at the expense of much longer response times, which might reduce the systems practicality in applied settings.
Furthermore, it is important to consider that the tests in this research consisted of only 50 facial pairs, and some super-recognisers (n = 31 out of 317; 10.2%), and above-average participants (n = 10 out of 841; 1.8%) but no average ability controls achieved maximum scores of 50 out of 50. In the workplace, identity verifiers may make very large numbers of decisions. Future research should therefore attempt to address this by increasing the number of trials to reduce celling effects and also use more difficult stimuli to replicate the challenges of real life.
Once again, a big thank you to all of the participants who completed in both experiments, your efforts remain essential to the continuation of our research and to all of the research team who helped with the completion of this project.
Group criteria
In this research, participants were classified into the three ability groups based on previous scores on the 102-trial Cambridge Face Memory Test: Extended (Russell et al., 2009) and the 40-trial Glasgow Face Matching Test (Burton et al., 2010).
References
Bobak, A. K., Hancock, P. J., & Bate, S. (2016). Super‐recognisers in action: Evidence from face‐matching and face memory tasks. Applied Cognitive Psychology, 30(1), 81-91. doi: 10.1002/acp.3170.
Burton, A. M., White, D., & McNeill, A. (2010). The Glasgow face matching test. Behavior Research Methods, 42, 286–291. doi:10.3758/BRM.42.1.286.
Robertson, D. J., Fysh, M. C., & Bindemann, M. (2019). Face identity verification: Five challenges facing practitioners. Keesing Journal of Documents & Identity, 59, 3-8. ISSN 1871-272X.
Russell, R., Duchaine, B., & Nakayama, K., (2009). Super-recognizers: People with extraordinary face recognition ability. Psychonomic Bulletin & Review, 16, 252–257. doi:10.3758/PBR.16.2.252
Towler, A., White, D., & Kemp, R. I. (2017). Evaluating the feature comparison strategy for forensic face identification. Journal of Experimental Psychology: Applied, 23(1), 47-58. doi: 10.1037/xap0000108.
Weatherford, D. R., Erickson, W. B., Thomas, J., Walker, M. E., & Schein, B. (2020). You shall not pass: how facial variability and feedback affect the detection of low-prevalence fake IDs. Cognitive Research: Principles and Implications, 5(1), 1-15. doi: 10.1186/s41235-019-0204-1.
Comments