From facial images of different quality to score based LR

A simple method is proposed to assess the quality of a trace facial image in the context of the facial recognition system used using the similarity scores with low quality different-source facial images, defined as the Confusion Score (CS). Methods are proposed to calculate the probability of finding the correct facial image in a database using low quality images for investigational purposes using the CS, as well as calculation of the Likelihood Ratio (LR) for comparison of low quality trace facial images with good quality reference facial images, based on the assessed CS of the trace image. Improvement of performance of an LR-system using training datasets stratified on CS over the use of pooled data is demonstrated. Examples of using the proposed approach in simulated case examples are presented. •A simple method to measure the quality of facial images using Confusion Scores (CS).•Calculation of the probability of finding the correct facial image in a database.•Calculation of the score-based Likelihood Ratio (LR) for comparison.•Improvement of performance of a score-based LR-system by CS stratification.