The 89th Annual Meeting of the American Association of Physical Anthropologists (2020)

Reliability of facial identification methods in optimal photographs and suboptimal CCTV footage


Human Variation and Identification Research Unit, School of Anatomical Sciences, University of the Witwatersrand

April 18, 2020 4:15PM, Diamond 8-9 Add to calendar

Facial comparison (FC) is routinely utilized in the medicolegal context through holistic review and morphological analysis. These methods have, to date, not been systematically validated, particularly when involving CCTV recordings. Increased global availability of video surveillance data emphasizes the need for accurate facial comparison methods. This study aimed to test current FC methodologies on a photographic and video recording sample.

Face pools (FP) of 175 African males of facial photographs in multiple views were compiled. Each face pool included one target photograph or CCTV recording and 10 potential matching photos. Photos of the first 75 participants were taken in standardized settings and compared to other suboptimal photos in a photo-to-photo cohort. The remaining FP (n=100) included CCTV recordings of participants compared to standardized photographs. FP were firstly analyzed via holistic review to identify potential matches. Potentially corresponding faces were then tested via morphological analysis. Method accuracy was determined through hit-rate, sensitivity and specificity calculation.

Higher accuracy for morphological analysis (κ=0.96, p<0.01) compared to holistic review (κ=0.76, p<0.01) was found within the photo-to-photo cohort. High sensitivity was identified for both methods (1.00 and 0.97), while a slightly higher specificity was identified for morphological analysis (0.99) over holistic review (0.95). Preliminary CCTV results show decreased accuracies for holistic (κ=0.65, p<0.01) and morphological (κ=0.88, p<0.01) analyses.

In conclusion, morphological analysis, although subjective, yielded better results. Overall higher accuracies were observed within the photo-to-photo cohort compared to CCTV recordings as it involves higher quality, controlled images.

The National Research Foundation of South Africa sponsored this research. Any information expressed in this study is the authors’ and therefore the NRF does not accept any liability in regards.