Assessing how facial recognition systems perform with operational imagery and how this affects decision-making.

Facial recognition systems can achieve high levels of accuracy under controlled conditions, but performance decreases when applied to real-world imagery, such as CCTV.

This project involved working with a UK law enforcement agency to evaluate a commercial facial recognition system using operationally relevant data, including CCTV and custody images.

The focus was on assessing how system performance changes under different image conditions, and how this affects the interpretation of candidate matches returned by the system.

The work also highlights how system performance can be misinterpreted if not assessed in the context of real-world conditions.

Statistical analysis was used to benchmark performance and to develop metrics that support more reliable and informed decision-making when reviewing search results.