Automated face recognition in forensic science: Review and perspectives

Area of Science:

• Forensic Science
• Computer Science
• Biometrics

Background:

• Forensic face recognition is increasingly used in legal investigations and evidence presentation.
• Current automatic systems lack methodological standardization and empirical validation, impacting their courtroom reliability.
• There is a need for more studies using forensic data to establish reliable evidence-based practices.

Purpose of the Study:

• To review existing literature on forensic face recognition.
• To establish a methodological workflow for a score-based likelihood ratio (LR) computation model using a Bayesian framework.
• To address the need for standardized and validated methods for presenting automated face comparison results in court.

Main Methods:

• Literature review to establish a methodological workflow.
• Development of a score-based likelihood ratio computation model within a Bayesian framework.
• Exploration of different approaches for modeling within-source and between-source variability distributions.
• Assessment of model performance using discriminating power and calibration metrics.

Main Results:

• Proposed a Bayesian framework for score-based LR computation in forensic face recognition.
• Discussed case-specific versus generic modeling approaches and their legal defensibility.
• Highlighted the importance of assessing discriminating power and calibration for model robustness.
• Described key performance metrics like Equal Error Rate (EER) and Cost of log likelihood-ratio (CostL L R), and graphical tools such as Tippett plots.

Conclusions:

• The proposed workflow and Bayesian framework aim to enhance the reliability and standardization of forensic face recognition evidence.
• Robust performance assessment using defined metrics and visualizations is crucial for validating these systems for court use.
• Further empirical studies with forensic data are essential to solidify the role of automated face recognition in the judicial system.