Evaluating match confidence in automated face recognition via likelihood ratio determination: A case study

Area of Science:

• Forensic Science
• Computer Science
• Statistical Analysis

Background:

• Forensic facial examination (FFE) traditionally relies on manual comparison by experts.
• Automated facial recognition (AFR) uses AI for suspect identification but requires expert validation.
• Current methods lack robust statistical frameworks for court admissibility.

Purpose of the Study:

• To develop an innovative FFE workflow integrating AFR output with Bayesian statistical analysis.
• To provide court-admissible statistical results for facial comparisons.
• To enhance the reliability and legal accountability of facial identification.

Main Methods:

• Generation of facial image match scores using automated software.
• Calculation of likelihood ratios (LRs) via kernel density estimation (KDE) and Bayesian statistical modeling.
• Validation using Tippett Plots aligned with European Network of Forensic Science Institutes (ENFSI) guidelines.

Main Results:

• The proposed framework generates court-admissible statistical results for facial comparisons.
• The methodology ensures forensic reliability through rigorous statistical validation.
• Practitioner oversight is maintained within the enhanced workflow.

Conclusions:

• The innovative FFE workflow effectively integrates AFR with Bayesian statistics for reliable facial identification.
• This approach supports forensic practitioners with statistically sound evidence for courtroom presentation.
• The methodology has been shared with ENFSI experts, indicating potential for wider adoption.